Copying More Repo to Demo

This commit is contained in:
GAJ3T 2026-10-26 05:16:59 -04:00
parent 8785dcf45d
commit cb66dcb8e3
11 changed files with 2054 additions and 0 deletions

5
AceStepXL-Training/.gitignore vendored Normal file
View File

@ -0,0 +1,5 @@
workspace/
__pycache__/
*.py[cod]
.venv/
.env

View File

@ -0,0 +1,225 @@
# Voice Studio
A local, browser-based voice-character training workbench for ACE-Step 1.5 XL.
Maintain any number of characters, each with separate recordings, annotations,
dataset versions, jobs, logs, and exports.
## Delivery status
Source implementation only. At the user's request, the application has **not**
been launched or tested, dependencies have not been installed, and no models
have been trained. No services or containers have been started.
The default paths reflect the existing local installation inspected during
development. Header inspection found six stereo, 48 kHz PCM-16 WAV recordings
in the initial character folder, totaling approximately 17 minutes. The audio
has not been listened to or assessed for vocal quality or training permission.
## Start when ready
From this project directory in PowerShell:
```powershell
.\Start-VoiceStudio.ps1
```
Open <http://127.0.0.1:9874>. The launcher does not install packages or open
another ACE-Step instance. It uses:
- ACE-Step installation: `D:\ai\ACE-Step-1.5`
- Python: `D:\ai\ACE-Step-1.5\.venv\Scripts\python.exe`
- Character library: `..\Voice Characters`
- Generated data: `.\workspace`
To supply another interpreter or installation:
```powershell
.\Start-VoiceStudio.ps1 -AceStepRoot "D:\Tools\ACE-Step-1.5" `
-Python "D:\Tools\ACE-Step-1.5\.venv\Scripts\python.exe" -Port 9874
```
The browser server requires Python 3.10+ and the small set of packages in
`requirements.txt`. It does not install the ACE-Step runtime. If those
packages are missing, install them deliberately into the chosen interpreter:
```powershell
& "D:\ai\ACE-Step-1.5\.venv\Scripts\python.exe" -m pip install -r .\requirements.txt
```
Review changes to a shared environment before installing. ACE-Step itself
must have its own CUDA-enabled Torch, training dependencies, and downloaded
model assets. Training requires an NVIDIA CUDA GPU; memory requirements
depend on the XL variant, clip duration, and upstream implementation.
No Docker, Postman, cloud services, paid APIs, or MCP server is required.
This is a local web application, not an MCP integration. The browser server
binds only to `127.0.0.1`; do not expose it through a public proxy.
## Multiple voice characters
Every immediate subfolder of the character library is a separate character.
The existing `GothMetalMaleV2` folder is discovered, not hard-coded.
Use **New character** to create another folder and independent workspace,
then **Add recordings** to copy WAV, FLAC, or OGG files into it. Uploads receive
unique filenames; existing files are never overwritten. Alternatively, add
a character subfolder and audio files directly to the library before opening
or refreshing the studio.
Display names, generation triggers, default captions, and language are
editable. Use a distinct trigger per voice. Identity changes apply to future
datasets and do not silently rewrite old annotations.
Environment overrides:
| Variable | Meaning |
|---|---|
| `ACESTEP_ROOT` | Default ACE-Step installation |
| `VOICE_STUDIO_SOURCE` | Default recording library |
| `VOICE_STUDIO_WORKSPACE` | Generated datasets, settings, job history, and exports |
Saved settings override the default installation and source paths.
The output workspace is selected before launch, not moved through the UI.
## Dataset preparation and review
1. Select or create a character.
2. Add **isolated vocals** that you own or are authorized to use.
3. Set the clip duration and silence threshold, then prepare a dataset.
4. Listen to the resulting clips. Correct captions and enter accurate lyrics,
syllables, or descriptions of the actual non-lexical vocalisation.
5. Mark each clip reviewed and save.
Preparation creates a new version every time:
```text
workspace\
<character-id>\
character.json
datasets\
<version>\
ace_audio\ 48 kHz stereo PCM-16 WAV
svc_audio\ 40 kHz mono PCM-16 WAV
dataset.json Native ACE-Step samples/metadata
segments.json Original paths and segment timestamps
tensors\ Generated only by explicit preprocessing
ready.json Written only after complete preprocessing
jobs\
<job-id>\
job.json
output.log
recipe.json Training jobs only
adapter\ Training outputs, if produced by the backend
```
Silence detection uses 20 ms energy windows, brief boundary padding, and a
minimum retained segment length of 1.5 seconds. Long spans are cut near a
low-energy boundary where possible; audition cuts because this is not a
phoneme-aware segmenter. Quiet clips below the configured threshold are
excluded and reported. Very short isolated sounds may be omitted.
Audio is not denoised, separated, pitch-corrected, or transcribed automatically.
Resampling peaks are limited only when needed to prevent output clipping.
For SVC, a phase-cancelling stereo downmix falls back to the strongest channel
and produces a warning. Original recordings are never modified.
Unknown lyrics remain blank. Vocal data is explicitly **not** marked
instrumental. ACE-Step preprocessing is blocked until every clip has vocal
text, a caption, and a listening-review confirmation. This does not guarantee
that an annotation is correct; it is a human review gate.
## Path A: new songs with ACE-Step XL
1. Open **Train a voice**.
2. Select the exact installed XL variant: XL-SFT, XL-Turbo, or XL-Base.
3. Confirm permission and suitable vocal data.
4. Run **Preprocess clips**. The studio invokes upstream's two-pass
`preprocess_audio_files` in a separate process using the ACE-Step interpreter.
5. Run **Start training** after preprocessing completes.
The integration targets the local ACE-Step source interfaces inspected at
commit `14c0211`:
- `acestep.training_v2.preprocess.preprocess_audio_files`
- `train.py --plain --yes fixed`
Commands use the literal checkpoint directory name plus an explicit
`--base-model xl_*`. This avoids the inspected CLI path validator's missing
XL aliases while retaining the correct XL family for training configuration.
No upstream files are modified.
Starting settings: batch size 1, BF16, rank 16, learning rate 0.0001,
gradient accumulation 4, gradient checkpointing, encoder offload, AdamW,
and zero DataLoader workers. These are starting settings, not a validated
24 GB memory guarantee. Default 12-second clips reduce activation memory.
Close other GPU workloads before training. Lower clip length/rank if needed.
Every preprocessing attempt gets a separate tensor directory. Completion
requires all expected samples to succeed. A dataset hash and XL variant are
recorded; changes invalidate readiness and require preprocessing again.
The console displays actual subprocess output, not estimated loss curves or
invented progress. Nonzero exits and explicit backend `[FAIL]` reports are
failures. Training completion additionally requires final `.safetensors`
weights and their adapter configuration. That check does **not** certify voice
quality: audition checkpoints with prompts and melodies not used in training.
Load the resulting adapter **directory**, retaining `adapter_config.json` and
its weights together, through the matching ACE-Step XL model's adapter loader.
Include the trigger recorded in that training job's `recipe.json` in the
generation caption; renaming a character does not rewrite an old model's
trigger. In ComfyUI, the
installed nodes must explicitly support the XL variant and the saved adapter
format; compatibility is not assumed and weights are not silently converted.
Syllable runs can provide timbre and pitch coverage, but do not establish good
lyric pronunciation or full-song generalisation. A voice-only set can also
bias outputs toward a cappella performances. Add authorized clean sung phrases
with varied pitches, vowels, consonants, dynamics, and accurately annotated
lyrics; evaluate on held-out performances before relying on the character.
## Path B: singing voice conversion
For preserving an existing singer's melody, timing, and performance while
changing the voice, use a **separate RVC model**, for example through
[Applio](https://github.com/IAHispano/Applio).
1. Prepare and listen to the character's vocal clips.
2. Open **Models & exports** and build an SVC dataset export.
3. Download and extract the ZIP. It contains the mono 40 kHz WAVs and local
training handoff instructions, not a trained conversion model.
4. In a separately installed Applio environment, select a unique model name
and the extracted character folder; use 40 kHz, preprocess, extract
pitch/features, train, and build its retrieval index following the current
Applio training guide.
5. Use its `.pth` and `.index` for conversion of isolated source vocals, then
mix the converted vocals back with the instrumental.
Applio installation, downloads, training, and conversion are **not automated**
by this studio. That boundary avoids guessing an unselected backend version
or mixing conflicting Torch dependencies. SVC training generally does not
need lyrics, so the export is available without ACE-Step lyric annotation.
It still requires human listening and authorized audio.
An ACE-Step LoRA is not an RVC/so-vits-svc model. Renaming an adapter does not
convert it. ACE-Step audio conditioning/cover workflows may be useful, but
are not a guaranteed drop-in replacement for performance-preserving SVC.
## Job and file behaviour
- One dataset/GPU/export job runs at a time. Mutations are blocked during jobs.
- History, logs, and artifacts persist per character. Interrupted jobs are
marked failed on the next studio startup, never reported as completed.
- Stopping the studio requests termination of its active backend process.
There is no browser cancellation/resume UI. A subsequent training request
starts a new run; native CLI resume remains an upstream advanced workflow.
- Preparation and uploads write only new files. Earlier datasets and partial
failed runs are retained for inspection; no automatic deletion is performed.
- Model weights and audio remain local. The application contains no telemetry,
remote fonts, external JavaScript, or audio-upload services.
### Upstream references
- [ACE-Step 1.5](https://github.com/ace-step/ACE-Step-1.5)
- [ACE-Step LoRA training tutorial](https://github.com/ace-step/ACE-Step-1.5/blob/main/docs/en/LoRA_Training_Tutorial.md)
- [Applio](https://github.com/IAHispano/Applio)

View File

@ -0,0 +1,29 @@
param(
[string]$AceStepRoot = "D:\ai\ACE-Step-1.5",
[string]$Python = "",
[ValidateRange(1024, 65535)]
[int]$Port = 9874
)
$ErrorActionPreference = "Stop"
if ($env:ACESTEP_ROOT -and -not $PSBoundParameters.ContainsKey("AceStepRoot")) {
$AceStepRoot = $env:ACESTEP_ROOT
}
if (-not $Python) { $Python = Join-Path $AceStepRoot ".venv\Scripts\python.exe" }
if (-not (Test-Path -LiteralPath $Python -PathType Leaf)) {
throw "Python not found: $Python. Supply -Python with a prepared Python interpreter."
}
$env:ACESTEP_ROOT = $AceStepRoot
$env:PYTHONPATH = Join-Path $PSScriptRoot "src"
if (-not $env:VOICE_STUDIO_SOURCE) {
$env:VOICE_STUDIO_SOURCE = Join-Path (Split-Path $PSScriptRoot -Parent) "Voice Characters"
}
Write-Host ""
Write-Host "VOICE STUDIO" -ForegroundColor Magenta
Write-Host "Open http://127.0.0.1:$Port in your browser."
Write-Host "Training starts only when you explicitly request it in the studio."
Write-Host "Press Ctrl+C to stop the studio."
Write-Host ""
& $Python -m voice_studio.server --port $Port
exit $LASTEXITCODE

View File

@ -0,0 +1,6 @@
fastapi>=0.115,<1
uvicorn>=0.30,<1
python-multipart>=0.0.18,<1
numpy>=1.26,<3
scipy>=1.12,<2
soundfile>=0.12,<1

View File

@ -0,0 +1 @@
"""Local voice-character dataset and training studio."""

View File

@ -0,0 +1,47 @@
"""Isolate GPU-heavy upstream preprocessing from the browser server."""
from __future__ import annotations
import argparse
import json
import logging
import sys
from pathlib import Path
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--ace-root", type=Path, required=True)
parser.add_argument("--dataset", type=Path, required=True)
parser.add_argument("--output", type=Path, required=True)
parser.add_argument("--model", required=True)
args = parser.parse_args()
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
sys.path.insert(0, str(args.ace_root.resolve()))
from acestep.training_v2.preprocess import preprocess_audio_files
data = json.loads(args.dataset.read_text(encoding="utf-8"))
expected = len(data["samples"])
if not expected:
raise ValueError("The dataset is empty.")
for sample in data["samples"]:
if not sample.get("reviewed") or not sample.get("lyrics", "").strip():
raise ValueError(f"Clip requires reviewed lyrics: {sample['filename']}")
if not Path(sample["audio_path"]).is_file():
raise FileNotFoundError(sample["audio_path"])
result = preprocess_audio_files(
audio_dir=None, output_dir=str(args.output),
checkpoint_dir=str(args.ace_root / "checkpoints"), variant=args.model,
max_duration=max(sample["duration"] for sample in data["samples"]) + 1,
dataset_json=str(args.dataset), device="cuda", precision="bf16",
progress_callback=lambda current, total, message: print(
f"[{current}/{total}] {message}", flush=True
),
)
if result["failed"] or result["processed"] != expected or result["total"] != expected:
raise RuntimeError(f"Incomplete preprocessing: expected {expected}, received {result}")
print(json.dumps(result), flush=True)
if __name__ == "__main__":
main()

View File

@ -0,0 +1,299 @@
from __future__ import annotations
import argparse
import os
import subprocess
import uuid
from contextlib import asynccontextmanager
from pathlib import Path
from typing import Literal
from urllib.parse import urlsplit
import soundfile as sf
from fastapi import FastAPI, File, HTTPException, Request, UploadFile
from fastapi.responses import FileResponse, JSONResponse
from fastapi.staticfiles import StaticFiles
from pydantic import BaseModel, Field
from .studio import (
MODELS, WORKSPACE, character, characters, create_character, dataset_path, dataset_view,
jobs, read_json, save_character, settings, slug, training_command, write_json,
)
STATIC = Path(__file__).with_name("static")
@asynccontextmanager
async def lifespan(app: FastAPI):
WORKSPACE.mkdir(parents=True, exist_ok=True)
jobs.load_history()
yield
jobs.shutdown()
app = FastAPI(title="Voice Studio", docs_url=None, redoc_url=None, lifespan=lifespan)
@app.middleware("http")
async def local_only(request: Request, call_next):
host = urlsplit("//" + request.headers.get("host", "")).hostname
origin = request.headers.get("origin")
if host not in {"localhost", "127.0.0.1", "::1"}:
return JSONResponse({"detail": "Voice Studio accepts local connections only."}, status_code=403)
if origin and origin != f"{request.url.scheme}://{request.headers.get('host')}":
return JSONResponse({"detail": "Cross-origin requests are not allowed."}, status_code=403)
response = await call_next(request)
response.headers["X-Content-Type-Options"] = "nosniff"
response.headers["Referrer-Policy"] = "no-referrer"
response.headers["Cache-Control"] = "no-store"
return response
@app.exception_handler(ValueError)
async def value_error(request: Request, exc: ValueError):
return JSONResponse({"detail": str(exc)}, status_code=400)
@app.exception_handler(OSError)
async def io_error(request: Request, exc: OSError):
return JSONResponse({"detail": str(exc)}, status_code=404 if isinstance(exc, FileNotFoundError) else 400)
class CharacterInput(BaseModel):
name: str = Field(min_length=1, max_length=80)
trigger: str = Field(default="", max_length=64)
class ProfileInput(BaseModel):
name: str = Field(min_length=1, max_length=80)
trigger: str = Field(min_length=1, max_length=64)
caption: str = Field(min_length=1, max_length=2000)
language: str = Field(min_length=1, max_length=32)
class SettingsInput(BaseModel):
ace_root: str = Field(min_length=1)
ace_python: str = Field(min_length=1)
source_root: str = Field(min_length=1)
class AnnotationInput(BaseModel):
dataset_version: str = Field(min_length=1, max_length=80)
caption: str = Field(min_length=1, max_length=4000)
lyrics: str = Field(default="", max_length=10000)
reviewed: bool = False
class JobInput(BaseModel):
action: Literal["prepare", "preprocess", "train", "export_svc"]
model: Literal[
"acestep-v15-xl-sft", "acestep-v15-xl-turbo", "acestep-v15-xl-base"
] = "acestep-v15-xl-sft"
clip_seconds: float = Field(default=12, ge=3, le=30)
silence_db: float = Field(default=-45, ge=-70, le=-20)
epochs: int = Field(default=100, ge=1, le=2000)
rank: Literal[8, 16, 32, 64] = 16
learning_rate: float = Field(default=0.0001, ge=0.000001, le=0.001)
accumulation: int = Field(default=4, ge=1, le=32)
authorized: bool = False
def model_status(config: dict) -> list[dict]:
return [
{"id": model, "label": model.replace("acestep-v15-", "").upper().replace("-", " "),
"available": (Path(config["ace_root"]) / "checkpoints" / model / "config.json").is_file()}
for model in MODELS
]
def artifacts(cid: str) -> list[dict]:
base = WORKSPACE / cid
result = []
if not base.is_dir():
return result
for path in base.rglob("*"):
if not path.is_file():
continue
if path.suffix not in {".zip", ".safetensors"} and path.name not in {
"adapter_config.json", "recipe.json", "dataset.json", "segments.json", "output.log",
}:
continue
result.append({
"name": path.name, "relative": str(path.relative_to(WORKSPACE)).replace(os.sep, "/"),
"bytes": path.stat().st_size, "modified": path.stat().st_mtime,
})
return sorted(result, key=lambda item: item["modified"], reverse=True)
@app.get("/api/bootstrap")
def bootstrap():
config = settings()
return {
"settings": config, "characters": characters(), "models": model_status(config),
"jobs": jobs.snapshot(), "busy": jobs.busy,
"backend_available": Path(config["ace_python"]).is_file()
and (Path(config["ace_root"]) / "train.py").is_file(),
"source_available": Path(config["source_root"]).is_dir(),
}
@app.post("/api/characters", status_code=201)
def add_character(body: CharacterInput):
with jobs.lock:
jobs.ensure_idle()
return create_character(body.name, body.trigger)
@app.get("/api/characters/{cid}")
def get_character(cid: str):
item = character(cid)
return {**item, "prepared": dataset_view(item), "artifacts": artifacts(cid)}
@app.put("/api/characters/{cid}")
def update_character(cid: str, body: ProfileInput):
with jobs.lock:
jobs.ensure_idle()
item = character(cid)
item.update(body.model_dump())
item["trigger"] = slug(body.trigger)
save_character(item)
return item
@app.post("/api/characters/{cid}/recordings")
def upload_recordings(cid: str, files: list[UploadFile] = File(...)):
with jobs.lock:
jobs.ensure_idle()
item = character(cid)
accepted = []
for upload in files:
filename = (upload.filename or "").replace("\\", "/").rsplit("/", 1)[-1]
suffix = Path(filename).suffix.lower()
if suffix not in {".wav", ".flac", ".ogg"}:
raise ValueError("Supported recordings: WAV, FLAC, OGG. Earlier files in this batch may have been saved.")
target = Path(item["source"]) / f"{slug(Path(filename).stem)}-{uuid.uuid4().hex[:8]}{suffix}"
temporary = target.with_suffix(suffix + ".upload")
try:
size = 0
with temporary.open("xb") as handle:
while chunk := upload.file.read(1024 * 1024):
size += len(chunk)
if size > 512 * 1024 * 1024:
raise ValueError("Each recording must be under 512 MiB. Earlier uploads remain saved.")
handle.write(chunk)
try:
info = sf.info(str(temporary))
except (RuntimeError, sf.LibsndfileError) as exc:
raise ValueError(f"{filename}: unsupported or damaged audio: {exc}") from exc
if not info.frames or info.channels not in {1, 2}:
raise ValueError(f"{filename}: expected nonempty mono/stereo audio.")
temporary.replace(target)
accepted.append(target.name)
finally:
temporary.unlink(missing_ok=True)
upload.file.close()
return {"saved": accepted}
@app.get("/api/characters/{cid}/recordings/{rid}")
def recording_audio(cid: str, rid: str):
for recording in character(cid)["recordings"]:
if recording["id"] == rid:
return FileResponse(recording["path"])
raise HTTPException(404, "Recording not found.")
@app.get("/api/characters/{cid}/samples/{sid}/audio")
def sample_audio(cid: str, sid: str):
data = read_json(dataset_path(character(cid)))
for sample in data["samples"]:
if sample["id"] == sid:
path = Path(sample["audio_path"]).resolve()
if not path.is_relative_to(WORKSPACE / cid):
raise ValueError("Audio path is outside this character's workspace.")
return FileResponse(path, media_type="audio/wav")
raise HTTPException(404, "Clip not found.")
@app.put("/api/characters/{cid}/samples/{sid}")
def update_sample(cid: str, sid: str, body: AnnotationInput):
with jobs.lock:
jobs.ensure_idle()
path = dataset_path(character(cid))
if path.parent.name != body.dataset_version:
raise ValueError("The active dataset changed. Refresh the page before annotating its clips.")
data = read_json(path)
for sample in data["samples"]:
if sample["id"] == sid:
if body.reviewed and not body.lyrics.strip():
raise ValueError("Enter accurate lyrics or a description of the actual wordless vocalisation.")
sample.update(body.model_dump(exclude={"dataset_version"}))
sample.update(raw_lyrics=body.lyrics, labeled=body.reviewed, is_instrumental=False)
write_json(path, data)
return sample
raise HTTPException(404, "Clip not found.")
@app.post("/api/characters/{cid}/jobs", status_code=202)
def start_job(cid: str, body: JobInput):
return jobs.start(cid, body.action, body.model_dump())
@app.get("/api/jobs")
def get_jobs():
return {"jobs": jobs.snapshot(), "busy": jobs.busy}
@app.post("/api/characters/{cid}/recipe")
def preview_recipe(cid: str, body: JobInput):
item = character(cid)
prepared = dataset_view(item)
tensors = (Path(prepared["ready"]["tensor_dir"]) if prepared.get("ready")
else WORKSPACE / cid / "pending-tensors")
command = training_command(item, body.model_dump(), tensors, WORKSPACE / cid / "next-run" / "adapter")
return {"command": subprocess.list2cmdline(command), "args": command, "executed": False}
@app.put("/api/settings")
def update_settings(body: SettingsInput):
with jobs.lock:
jobs.ensure_idle()
config = body.model_dump()
if any(not Path(value).is_absolute() for value in config.values()):
raise ValueError("Use absolute local paths for all settings.")
if not Path(config["source_root"]).is_dir():
raise ValueError("The recording library folder must already exist.")
write_json(WORKSPACE / "settings.json", config)
return settings()
@app.get("/api/download/{relative:path}")
def download(relative: str):
path = (WORKSPACE / relative).resolve()
if not path.is_relative_to(WORKSPACE) or not path.is_file():
raise HTTPException(404, "Artifact not found.")
if path.suffix not in {".zip", ".safetensors", ".json", ".log"}:
raise HTTPException(400, "This file type is not available as an artifact.")
return FileResponse(path, filename=path.name)
@app.get("/")
def home():
return FileResponse(STATIC / "index.html")
app.mount("/static", StaticFiles(directory=str(STATIC)), name="static")
def main() -> None:
import uvicorn
parser = argparse.ArgumentParser(description="Voice Studio - local browser training workbench")
parser.add_argument("--port", type=int, default=9874)
args = parser.parse_args()
uvicorn.run(app, host="127.0.0.1", port=args.port, log_level="info")
if __name__ == "__main__":
main()

View File

@ -0,0 +1,500 @@
"use strict";
const $ = (selector) => document.querySelector(selector);
const state = {
characters: [], character: null, cid: null, sample: null, jobs: [], busy: false,
bootstrap: null, jobId: null, dirty: false, polling: false, toastTimer: null,
};
const pages = new Set(["overview", "recordings", "training", "exports", "settings"]);
const labels = {
prepare: "Dataset preparation", preprocess: "XL preprocessing",
train: "XL adapter training", export_svc: "SVC dataset export",
};
const escapeHTML = (value) => String(value ?? "").replace(/[&<>"']/g, (c) => ({
"&": "&amp;", "<": "&lt;", ">": "&gt;", '"': "&quot;", "'": "&#39;",
}[c]));
const icon = (name) => `<svg class="icon" aria-hidden="true"><use href="#i-${name}"/></svg>`;
const initials = (name) => name.replace(/([a-z])([A-Z])/g, "$1 $2").split(/[\s_-]+/)
.filter(Boolean).map((part) => part[0]).slice(0, 2).join("").toUpperCase() || "VC";
const duration = (seconds) => {
const total = Math.max(0, Math.round(Number(seconds) || 0));
return `${Math.floor(total / 60)}:${String(total % 60).padStart(2, "0")}`;
};
const size = (bytes) => bytes >= 1024 ** 3 ? `${(bytes / 1024 ** 3).toFixed(1)} GB`
: bytes >= 1024 ** 2 ? `${(bytes / 1024 ** 2).toFixed(1)} MB`
: `${Math.max(1, Math.round(bytes / 1024))} KB`;
const empty = (text) => `<div class="empty-state">${escapeHTML(text)}</div>`;
const downloadURL = (relative) => "/api/download/" + relative.split(/[\\/]/).map(encodeURIComponent).join("/");
async function api(path, options = {}) {
const init = { ...options };
if (init.body && !(init.body instanceof FormData)) {
init.headers = { "Content-Type": "application/json", ...init.headers };
init.body = JSON.stringify(init.body);
}
const response = await fetch("/api" + path, init);
const type = response.headers.get("content-type") || "";
if (!type.includes("application/json")) throw new Error(`Unexpected server response (${response.status}).`);
const result = await response.json();
if (!response.ok) {
const detail = Array.isArray(result.detail)
? result.detail.map((item) => `${item.loc?.join(".") || "Input"}: ${item.msg}`).join("; ")
: result.detail;
throw new Error(detail || `Request failed (${response.status}).`);
}
return result;
}
function notify(message, error = false) {
const toast = $("#toast");
toast.textContent = message;
toast.classList.toggle("error", error);
toast.classList.remove("hidden");
clearTimeout(state.toastTimer);
state.toastTimer = setTimeout(() => toast.classList.add("hidden"), error ? 12000 : 6000);
}
function report(error) {
console.error(error);
notify(error.message || String(error), true);
}
function changePage(page) {
if (!pages.has(page)) page = "overview";
document.querySelectorAll(".page").forEach((element) => {
element.classList.toggle("hidden", element.id !== `page-${page}`);
});
document.querySelectorAll(".nav-item").forEach((element) => {
element.classList.toggle("active", element.dataset.page === page);
});
if (location.hash !== `#${page}`) location.hash = page;
}
function canLeaveEdits() {
if (!state.dirty) return true;
if (!window.confirm("Discard the unsaved clip annotation?")) return false;
state.dirty = false;
return true;
}
async function refreshLibrary() {
const data = await api("/bootstrap");
state.bootstrap = data;
state.characters = data.characters;
state.jobs = data.jobs;
state.busy = data.busy;
const indicator = $("#backend-indicator");
indicator.classList.toggle("ready", data.backend_available);
indicator.replaceChildren();
indicator.append(document.createElement("i"), document.createTextNode(
data.backend_available ? "Local backend configured" : "Configure ACE-Step path",
));
$("#connection-error").classList.toggle("hidden", data.source_available);
if (!data.source_available) {
$("#connection-error").textContent = "The recording library folder was not found. Update its path in Workspace settings.";
}
$("#settings-ace").value = data.settings.ace_root;
$("#settings-python").value = data.settings.ace_python;
$("#settings-source").value = data.settings.source_root;
$("#settings-workspace").value = data.settings.workspace;
const oldModel = $("#model-select").value;
$("#model-select").innerHTML = data.models.map((model) =>
`<option value="${escapeHTML(model.id)}">${escapeHTML(model.label)}${model.available ? "" : " - not installed"}</option>`,
).join("");
if (data.models.some((model) => model.id === oldModel)) $("#model-select").value = oldModel;
if (!state.characters.some((character) => character.id === state.cid)) {
state.cid = state.characters[0]?.id || null;
state.sample = null;
}
renderLibrary();
}
function renderLibrary() {
$("#character-list").innerHTML = state.characters.map((character) => `
<button class="character-item ${character.id === state.cid ? "selected" : ""}" data-character="${character.id}">
<span class="avatar">${escapeHTML(initials(character.name))}</span>
<span class="character-text"><strong>${escapeHTML(character.name)}</strong>
<small>${character.recordings.length} recordings &middot; ${duration(character.duration)}</small></span>
</button>`).join("") || empty("Your characters will appear here.");
$("#library-total").textContent = `${state.characters.length} CHARACTER${state.characters.length === 1 ? "" : "S"}`;
$("#library-grid").innerHTML = state.characters.map((character) => `
<button class="library-card" data-character="${character.id}">
<span class="avatar">${escapeHTML(initials(character.name))}</span>
<span><h3>${escapeHTML(character.name)}</h3><p>${character.recordings.length} recordings &middot; ${duration(character.duration)}</p></span>
${icon("arrow")}
</button>`).join("") || empty("Create your first character to start building a voice library.");
}
async function refreshCharacter() {
state.character = state.cid ? await api(`/characters/${state.cid}`) : null;
renderCharacter();
renderJobs();
}
async function refreshAll() {
await refreshLibrary();
await refreshCharacter();
}
async function selectCharacter(cid) {
if (cid === state.cid || !canLeaveEdits()) return;
state.cid = cid;
state.sample = null;
state.jobId = null;
$("#authorized").checked = false;
$("#command-preview").textContent = "Your command preview will appear here.\nPreviewing does not execute anything.";
renderLibrary();
await refreshCharacter();
}
function renderCharacter() {
const item = state.character;
const prepared = item?.prepared;
const samples = prepared?.samples || [];
const recordings = item?.recordings || [];
const weights = (item?.artifacts || []).filter((artifact) => artifact.name.endsWith(".safetensors"));
$("#breadcrumb-name").textContent = item?.name || "Voice library";
$("#hero-name").textContent = item?.name || "A place for every voice.";
$("#nav-count").textContent = recordings.length;
$("#stat-recordings").textContent = recordings.length;
$("#stat-duration").textContent = duration(item?.duration);
$("#stat-clips").textContent = `${prepared?.reviewed || 0} / ${samples.length}`;
$("#stat-models").textContent = weights.length;
$("#export-trigger").textContent = prepared?.metadata?.custom_tag || item?.trigger || "No character selected";
$("#stage-source").textContent = recordings.length ? "RECORDINGS ADDED" : "ADD RECORDINGS";
$("#stage-source").classList.toggle("success", recordings.length > 0);
$("#stage-review").textContent = samples.length && prepared.reviewed === samples.length ? "CLIPS REVIEWED" : "PREPARE & REVIEW";
$("#stage-review").classList.toggle("success", samples.length > 0 && prepared.reviewed === samples.length);
$("#stage-train").textContent = weights.length ? "ADAPTER FILES SAVED" : "TRAIN WHEN READY";
$("#stage-train").classList.toggle("success", weights.length > 0);
$("#source-count").textContent = `${recordings.length} FILES`;
$("#source-path").textContent = item?.source || "Create a character to add recordings.";
$("#source-list").innerHTML = recordings.map((recording) => `
<div class="source-row"><span class="file-icon">${icon("wave")}</span>
<div class="source-details"><strong>${escapeHTML(recording.name)}</strong>
<small>${recording.error ? escapeHTML(recording.error) : `${duration(recording.duration)} &middot; ${recording.sample_rate / 1000} kHz &middot; ${recording.channels === 2 ? "Stereo" : "Mono"} &middot; ${escapeHTML(recording.subtype)}`}</small></div>
${recording.error ? "" : `<audio controls preload="none" aria-label="${escapeHTML(recording.name)}" src="/api/characters/${item.id}/recordings/${recording.id}"></audio>`}
</div>`).join("") || empty("No recordings yet. Add WAV, FLAC, or OGG files for this character.");
$("#profile-name").value = item?.name || "";
$("#profile-trigger").value = item?.trigger || "";
$("#profile-caption").value = item?.caption || "";
$("#profile-language").value = item?.language || "";
$("#review-count").textContent = samples.length
? `${prepared.reviewed} / ${samples.length} REVIEWED` : "NO DATASET YET";
const warnings = prepared?.warnings || [];
$("#dataset-warnings").classList.toggle("hidden", warnings.length === 0);
$("#dataset-warnings").textContent = warnings.join(" ");
if (!samples.some((sample) => sample.id === state.sample)) {
state.sample = samples.find((sample) => !sample.reviewed)?.id || samples[0]?.id || null;
}
renderClips();
if (!state.dirty) renderEditor();
renderArtifacts();
renderReadiness();
applyButtons();
}
function renderClips() {
const samples = state.character?.prepared?.samples || [];
$("#clip-list").innerHTML = samples.map((sample) => `
<button class="clip-item ${sample.id === state.sample ? "selected" : ""} ${sample.reviewed ? "reviewed" : ""}" data-sample="${escapeHTML(sample.id)}">
${icon(sample.reviewed ? "check" : "play")}<strong>${escapeHTML(sample.filename)}</strong>
<small>${sample.duration.toFixed(1)} s ${sample.reviewed ? "&middot; reviewed" : ""}</small>
</button>`).join("") || empty("Prepare a dataset to review its clips here.");
}
function renderEditor() {
const sample = state.character?.prepared?.samples.find((item) => item.id === state.sample);
$("#clip-heading").textContent = sample?.filename || "Choose a clip";
$("#clip-duration").textContent = sample ? `${sample.duration.toFixed(1)} SECONDS` : "--";
const audio = $("#clip-audio");
audio.pause();
if (sample) {
audio.src = `/api/characters/${state.cid}/samples/${encodeURIComponent(sample.id)}/audio`;
} else {
audio.removeAttribute("src");
}
$("#clip-caption").value = sample?.caption || "";
$("#clip-lyrics").value = sample?.lyrics || "";
$("#clip-reviewed").checked = Boolean(sample?.reviewed);
applyButtons();
}
function renderArtifacts() {
const artifacts = state.character?.artifacts || [];
$("#artifact-list").innerHTML = artifacts.map((artifact) => `
<div class="artifact-row"><span class="stat-icon ${artifact.name.endsWith(".zip") ? "mint" : "lavender"}">${icon("box")}</span>
<div class="artifact-detail"><strong>${escapeHTML(artifact.name)}</strong>
<small>${escapeHTML(artifact.relative)}</small></div>
<span class="artifact-size">${size(artifact.bytes)}</span>
<a class="button secondary" href="${downloadURL(artifact.relative)}" download>Download</a>
</div>`).join("") || empty("No artifacts saved yet. Prepared datasets, job logs, SVC exports, and trained adapters will appear here.");
}
function renderReadiness() {
const data = state.bootstrap;
const prepared = state.character?.prepared;
const samples = prepared?.samples || [];
const model = $("#model-select").value;
const checks = [
[Boolean(data?.backend_available), "ACE-Step runtime path configured"],
[Boolean(data?.models.find((item) => item.id === model)?.available), "Selected XL checkpoint present"],
[samples.length > 0, `${samples.length} clips in the prepared dataset`],
[samples.length > 0 && prepared.reviewed === samples.length, `${prepared?.reviewed || 0} clips listened to and annotated`],
[Boolean(prepared?.ready && prepared.ready.model === model), "Tensors match this dataset and XL variant"],
[$("#authorized").checked, "Voice-use authorization confirmed"],
];
$("#readiness-list").innerHTML = checks.map(([complete, text], index) => `
<li class="${complete ? "complete" : ""}"><span class="check-circle">${complete ? icon("check") : index + 1}</span>${escapeHTML(text)}</li>`,
).join("");
const complete = checks.filter(([done]) => done).length;
$("#readiness-pill").textContent = `${complete} / ${checks.length} READY`;
$("#readiness-pill").classList.toggle("success", complete === checks.length);
}
function applyButtons() {
document.querySelectorAll(".requires-character").forEach((button) => {
button.disabled = !state.cid || state.busy;
});
document.querySelectorAll(".add-character").forEach((button) => { button.disabled = state.busy; });
$("#profile-form button").disabled = !state.cid || state.busy;
$("#settings-form button").disabled = state.busy;
$("#save-clip").disabled = !state.sample || state.busy;
$("#next-clip").disabled = !state.sample || state.busy;
document.querySelectorAll("#annotation-form input, #annotation-form textarea").forEach((field) => {
field.disabled = !state.sample || state.busy;
});
document.querySelectorAll("#profile-form input").forEach((field) => {
field.disabled = !state.cid || state.busy;
});
}
function renderJobs() {
const filtered = state.jobs.filter((job) => job.character_id === state.cid);
if (!filtered.some((job) => job.id === state.jobId)) state.jobId = filtered[0]?.id || null;
$("#job-select").innerHTML = filtered.map((job) => `
<option value="${escapeHTML(job.id)}"${job.id === state.jobId ? " selected" : ""}>${escapeHTML(labels[job.action] || job.action)} - ${escapeHTML(job.status)}</option>`,
).join("") || '<option value="">No jobs yet</option>';
const job = filtered.find((item) => item.id === state.jobId);
$("#job-title").textContent = job ? labels[job.action] || job.action : "Training console";
$("#job-status").textContent = job?.status.toUpperCase() || "IDLE";
$("#job-status").className = `pill ${job?.status === "failed" ? "failed" : ""}`;
$("#job-dot").className = `status-dot ${job?.status || ""}`;
const consoleElement = $("#job-log");
const atEnd = consoleElement.scrollHeight - consoleElement.scrollTop - consoleElement.clientHeight < 60;
consoleElement.textContent = job
? job.log.join("\n") + (job.error ? `\n\nERROR: ${job.error}` : "")
: "Ready when you are.\nNo jobs have been started for this character.";
if (atEnd) consoleElement.scrollTop = consoleElement.scrollHeight;
$("#job-time").textContent = job
? `${new Date(job.started_at).toLocaleString()}${job.ended_at ? ` - ${job.status}` : " - running"}`
: "No elapsed time to report";
applyButtons();
}
function jobOptions(action) {
const base = {
action, clip_seconds: Number($("#clip-seconds").value), silence_db: Number($("#silence-db").value),
};
if (["prepare", "export_svc"].includes(action)) return base;
if (!$("#training-form").reportValidity()) throw new Error("Check the training settings.");
return {
...base, model: $("#model-select").value,
epochs: Number($("#epochs").value), rank: Number($("#rank").value),
learning_rate: Number($("#learning-rate").value), accumulation: Number($("#accumulation").value),
authorized: $("#authorized").checked,
};
}
async function startJob(action) {
if (!state.cid) throw new Error("Create or select a character first.");
if (action === "prepare" && !canLeaveEdits()) return;
if (["train", "preprocess"].includes(action)) {
if (state.dirty) throw new Error("Save the current clip annotation before preprocessing or training.");
if (!$("#authorized").checked) throw new Error("Confirm voice-use authorization in the training recipe.");
if (!window.confirm(`Start ${labels[action].toLowerCase()} now? This launches a real GPU job in your ACE-Step environment.`)) return;
}
const job = await api(`/characters/${state.cid}/jobs`, { method: "POST", body: jobOptions(action) });
state.jobId = job.id;
state.jobs.unshift(job);
state.busy = true;
renderJobs();
changePage("training");
notify(`${labels[action]} queued. Follow its output in the console.`);
}
async function pollJobs() {
if (state.polling || !state.bootstrap) return;
state.polling = true;
try {
const data = await api("/jobs");
const previous = new Map(state.jobs.map((job) => [job.id, job.status]));
const finished = data.jobs.filter((job) =>
previous.has(job.id) && previous.get(job.id) !== job.status && ["completed", "failed"].includes(job.status),
);
state.jobs = data.jobs;
state.busy = data.busy;
if (state.bootstrap.source_available) $("#connection-error").classList.add("hidden");
renderJobs();
for (const job of finished) {
notify(job.status === "failed"
? `${labels[job.action]} failed: ${job.error}`
: `${labels[job.action]} completed.`, job.status === "failed");
}
if (finished.length && !state.dirty) await refreshAll();
} catch (error) {
$("#connection-error").textContent = `Connection to Voice Studio interrupted: ${error.message}`;
$("#connection-error").classList.remove("hidden");
} finally {
state.polling = false;
}
}
document.addEventListener("click", (event) => {
const page = event.target.closest("[data-page]");
if (page) changePage(page.dataset.page);
const character = event.target.closest("[data-character]");
if (character) selectCharacter(character.dataset.character).catch(report);
const clip = event.target.closest("[data-sample]");
if (clip && canLeaveEdits()) {
state.sample = clip.dataset.sample;
renderClips();
renderEditor();
}
if (event.target.closest(".add-character")) {
$("#dialog-error").textContent = "";
$("#character-form").reset();
$("#character-dialog").showModal();
$("#new-name").focus();
}
});
$("#close-dialog").addEventListener("click", () => $("#character-dialog").close());
$("#character-form").addEventListener("submit", async (event) => {
event.preventDefault();
const button = event.target.querySelector('[type="submit"]');
button.disabled = true;
try {
if (!canLeaveEdits()) return;
const result = await api("/characters", { method: "POST", body: {
name: $("#new-name").value.trim(), trigger: $("#new-trigger").value.trim(),
} });
state.cid = result.id;
state.sample = null;
state.jobId = null;
$("#authorized").checked = false;
$("#character-dialog").close();
await refreshAll();
changePage("recordings");
notify(`${result.name} has its own workspace. Add recordings to get started.`);
} catch (error) {
$("#dialog-error").textContent = error.message;
} finally {
button.disabled = false;
}
});
$("#profile-form").addEventListener("submit", async (event) => {
event.preventDefault();
try {
if (!canLeaveEdits()) return;
await api(`/characters/${state.cid}`, { method: "PUT", body: {
name: $("#profile-name").value.trim(), trigger: $("#profile-trigger").value.trim(),
caption: $("#profile-caption").value.trim(), language: $("#profile-language").value.trim(),
} });
await refreshAll();
notify("Character identity saved. Existing clip annotations were preserved.");
} catch (error) { report(error); }
});
$("#upload-button").addEventListener("click", () => $("#upload-input").click());
$("#upload-input").addEventListener("change", async (event) => {
const files = [...event.target.files];
if (!files.length) return;
const form = new FormData();
files.forEach((file) => form.append("files", file));
$("#upload-button").disabled = true;
try {
const result = await api(`/characters/${state.cid}/recordings`, { method: "POST", body: form });
await refreshAll();
notify(`${result.saved.length} recording(s) added. Prepare a new dataset to include them.`);
} catch (error) {
report(error);
await refreshAll().catch(report);
} finally {
event.target.value = "";
applyButtons();
}
});
$("#annotation-form").addEventListener("input", () => { state.dirty = true; });
$("#annotation-form").addEventListener("submit", async (event) => {
event.preventDefault();
try {
await api(`/characters/${state.cid}/samples/${state.sample}`, { method: "PUT", body: {
dataset_version: state.character.prepared.version,
caption: $("#clip-caption").value.trim(), lyrics: $("#clip-lyrics").value.trim(),
reviewed: $("#clip-reviewed").checked,
} });
state.dirty = false;
await refreshCharacter();
notify("Clip saved. Any changed annotations require fresh preprocessing.");
} catch (error) { report(error); }
});
$("#next-clip").addEventListener("click", () => {
if (!canLeaveEdits()) return;
const samples = state.character?.prepared?.samples || [];
const index = samples.findIndex((sample) => sample.id === state.sample);
state.sample = samples[(index + 1) % samples.length]?.id || null;
renderClips();
renderEditor();
});
$("#prepare-button").addEventListener("click", () => startJob("prepare").catch(report));
$("#preprocess-button").addEventListener("click", () => startJob("preprocess").catch(report));
$("#export-svc").addEventListener("click", () => startJob("export_svc").catch(report));
$("#training-form").addEventListener("submit", (event) => {
event.preventDefault();
startJob("train").catch(report);
});
$("#authorized").addEventListener("change", renderReadiness);
$("#model-select").addEventListener("change", renderReadiness);
$("#job-select").addEventListener("change", (event) => {
state.jobId = event.target.value;
renderJobs();
});
$("#preview-recipe").addEventListener("click", async () => {
try {
const result = await api(`/characters/${state.cid}/recipe`, { method: "POST", body: jobOptions("train") });
$("#command-preview").textContent = result.command;
} catch (error) { report(error); }
});
$("#refresh-artifacts").addEventListener("click", () => refreshCharacter().catch(report));
$("#settings-form").addEventListener("submit", async (event) => {
event.preventDefault();
try {
if (!canLeaveEdits()) return;
await api("/settings", { method: "PUT", body: {
ace_root: $("#settings-ace").value.trim(), ace_python: $("#settings-python").value.trim(),
source_root: $("#settings-source").value.trim(),
} });
$("#authorized").checked = false;
await refreshAll();
notify("Workspace paths saved.");
} catch (error) { report(error); }
});
window.addEventListener("hashchange", () => changePage(location.hash.slice(1)));
window.addEventListener("beforeunload", (event) => {
if (state.dirty) {
event.preventDefault();
event.returnValue = "";
}
});
changePage(location.hash.slice(1));
refreshAll().catch((error) => {
$("#connection-error").textContent = `Could not load the local workspace: ${error.message}. Start Voice Studio with its PowerShell launcher.`;
$("#connection-error").classList.remove("hidden");
report(error);
});
setInterval(pollJobs, 2500);

View File

@ -0,0 +1,139 @@
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="theme-color" content="#f5f4f9">
<title>Voice Studio - Character Training</title>
<link rel="stylesheet" href="/static/styles.css">
<script defer src="/static/app.js"></script>
</head>
<body>
<svg class="icon-library" aria-hidden="true" xmlns="http://www.w3.org/2000/svg">
<symbol id="i-wave" viewBox="0 0 24 24"><path d="M3 10v4m4-9v14m5-17v20m5-17v14m4-9v4"/></symbol>
<symbol id="i-grid" viewBox="0 0 24 24"><rect x="3" y="3" width="7" height="7" rx="2"/><rect x="14" y="3" width="7" height="7" rx="2"/><rect x="3" y="14" width="7" height="7" rx="2"/><rect x="14" y="14" width="7" height="7" rx="2"/></symbol>
<symbol id="i-mic" viewBox="0 0 24 24"><rect x="9" y="2" width="6" height="13" rx="3"/><path d="M5 10v2a7 7 0 0 0 14 0v-2M12 19v3m-4 0h8"/></symbol>
<symbol id="i-spark" viewBox="0 0 24 24"><path d="m12 3 2.5 6.5L21 12l-6.5 2.5L12 21l-2.5-6.5L3 12l6.5-2.5L12 3Z"/></symbol>
<symbol id="i-box" viewBox="0 0 24 24"><path d="m12 3 9 5v9l-9 5-9-5V8l9-5Zm0 9 9-4M12 12 3 8m9 4v10"/></symbol>
<symbol id="i-settings" viewBox="0 0 24 24"><path d="M4 7h16M4 17h16"/><circle cx="8" cy="7" r="3"/><circle cx="16" cy="17" r="3"/></symbol>
<symbol id="i-plus" viewBox="0 0 24 24"><path d="M12 5v14M5 12h14"/></symbol>
<symbol id="i-arrow" viewBox="0 0 24 24"><path d="M5 12h14m-6-6 6 6-6 6"/></symbol>
<symbol id="i-upload" viewBox="0 0 24 24"><path d="M12 16V3m-5 5 5-5 5 5M4 16v5h16v-5"/></symbol>
<symbol id="i-check" viewBox="0 0 24 24"><path d="m5 12 4 4L19 6"/></symbol>
<symbol id="i-play" viewBox="0 0 24 24"><path d="m8 4 12 8-12 8V4Z"/></symbol>
<symbol id="i-folder" viewBox="0 0 24 24"><path d="M3 7V4h6l3 3h9v13H3V7Z"/></symbol>
</svg>
<aside class="sidebar">
<a class="brand" href="#overview" aria-label="Voice Studio home">
<span class="brand-mark"><svg class="icon"><use href="#i-wave"/></svg></span>
<span>voice<span class="brand-light">studio</span><small>CHARACTER TRAINING</small></span>
</a>
<div class="workspace-label">WORKSPACE <span class="local-dot">LOCAL</span></div>
<nav class="navigation" aria-label="Main navigation">
<button class="nav-item active" data-page="overview"><svg class="icon"><use href="#i-grid"/></svg>Overview</button>
<button class="nav-item" data-page="recordings"><svg class="icon"><use href="#i-mic"/></svg>Dataset &amp; recordings<span id="nav-count" class="nav-count">0</span></button>
<button class="nav-item" data-page="training"><svg class="icon"><use href="#i-spark"/></svg>Train a voice</button>
<button class="nav-item" data-page="exports"><svg class="icon"><use href="#i-box"/></svg>Models &amp; exports</button>
</nav>
<div class="library-label"><span>YOUR CHARACTERS</span><button class="icon-button add-character" title="Add a voice character" aria-label="Add a voice character"><svg class="icon"><use href="#i-plus"/></svg></button></div>
<div id="character-list" class="character-list" aria-label="Voice characters"></div>
<button class="sidebar-add add-character"><svg class="icon"><use href="#i-plus"/></svg>New character</button>
<div class="sidebar-bottom">
<div class="privacy-card"><span class="privacy-icon"><svg class="icon"><use href="#i-folder"/></svg></span><div><strong>Your voice. Your machine.</strong><p>Audio stays on your computer.</p></div></div>
<button class="nav-item" data-page="settings"><svg class="icon"><use href="#i-settings"/></svg>Workspace settings</button>
<div class="sidebar-version">VOICE STUDIO <span>ACE-STEP 1.5 XL</span></div>
</div>
</aside>
<div class="main-shell">
<header class="topbar">
<div class="breadcrumb">Workspace <span>/</span> <strong id="breadcrumb-name">Voice library</strong></div>
<div class="topbar-right"><span id="backend-indicator" class="connection"><i></i>Checking workspace</span><span class="avatar mini">VS</span></div>
</header>
<main>
<div id="connection-error" class="notice danger hidden" role="alert"></div>
<section id="page-overview" class="page">
<div class="page-heading"><div><p class="eyebrow">THE VOICE BEHIND THE MUSIC</p><h1>Your next voice starts here.</h1><p class="muted">Build a character. Shape its sound. Make it your own.</p></div><button class="button primary add-character"><svg class="icon"><use href="#i-plus"/></svg>New character</button></div>
<div class="hero">
<div class="hero-content"><span class="pill hero-pill">VOICE CHARACTER WORKSPACE</span><h2 id="hero-name">A place for every voice.</h2><p>From your recordings to a recognisable vocal identity.<br>One library, two distinct training paths.</p><div class="hero-actions"><button class="button white" data-page="recordings">Prepare your dataset<svg class="icon"><use href="#i-arrow"/></svg></button><button class="button ghost-light" data-page="training">Explore training</button></div></div>
<div class="voice-art" aria-hidden="true"><div class="orbit orbit-one"></div><div class="orbit orbit-two"></div><div class="orbit orbit-three"></div><div class="voice-core"><svg><use href="#i-wave"/></svg></div><span class="art-label">CRAFTED FROM YOUR SOUND</span></div>
</div>
<div class="stat-grid">
<article class="stat"><span class="stat-icon lavender"><svg class="icon"><use href="#i-mic"/></svg></span><div><span class="stat-label">Source recordings</span><strong id="stat-recordings">0</strong><small>Original files preserved</small></div></article>
<article class="stat"><span class="stat-icon peach"><svg class="icon"><use href="#i-wave"/></svg></span><div><span class="stat-label">Source audio</span><strong id="stat-duration">0:00</strong><small>In this character's library</small></div></article>
<article class="stat"><span class="stat-icon mint"><svg class="icon"><use href="#i-check"/></svg></span><div><span class="stat-label">Reviewed clips</span><strong id="stat-clips">0 / 0</strong><small>Listening &amp; annotation</small></div></article>
<article class="stat"><span class="stat-icon blue"><svg class="icon"><use href="#i-box"/></svg></span><div><span class="stat-label">Adapter weight files</span><strong id="stat-models">0</strong><small>Actual saved artifacts</small></div></article>
</div>
<div class="section-heading"><div><h2>A clear path to your character</h2><p class="muted">No hidden steps. No automatic training.</p></div><button class="text-button" data-page="training">Open training<svg class="icon"><use href="#i-arrow"/></svg></button></div>
<div class="workflow-grid">
<article class="step-card"><span class="step-number">01</span><span id="stage-source" class="pill">ADD RECORDINGS</span><h3>Start with a clean voice</h3><p>Add isolated vocal takes. Keep accompaniment, heavy reverb, and other singers out of the training set.</p><button class="text-button" data-page="recordings">Manage recordings<svg class="icon"><use href="#i-arrow"/></svg></button></article>
<article class="step-card"><span class="step-number">02</span><span id="stage-review" class="pill">PREPARE &amp; REVIEW</span><h3>Give the details a listen</h3><p>Create short clips, check each one, and add accurate lyrics or wordless-vocal descriptions for ACE-Step.</p><button class="text-button" data-page="recordings">Review your clips<svg class="icon"><use href="#i-arrow"/></svg></button></article>
<article class="step-card"><span class="step-number">03</span><span id="stage-train" class="pill">TRAIN WHEN READY</span><h3>Make the character yours</h3><p>Train an XL adapter for new songs, or export a dataset for a separate RVC voice-conversion model.</p><button class="text-button" data-page="exports">See output formats<svg class="icon"><use href="#i-arrow"/></svg></button></article>
</div>
<div class="section-heading"><div><h2>Your character library</h2><p class="muted">Each voice has its own data, training history, and exports.</p></div><span id="library-total" class="pill">0 CHARACTERS</span></div>
<div id="library-grid" class="library-grid"></div>
</section>
<section id="page-recordings" class="page hidden">
<div class="page-heading"><div><p class="eyebrow">GOOD MODELS START WITH GOOD DATA</p><h1>Dataset &amp; recordings</h1><p class="muted">Originals stay untouched. Every prepared clip keeps its source timestamps.</p></div><button id="upload-button" class="button primary requires-character"><svg class="icon"><use href="#i-upload"/></svg>Add recordings</button><input id="upload-input" type="file" accept=".wav,.flac,.ogg" multiple hidden></div>
<div class="notice"><strong>Listen first.</strong> These tools split and resample audio; they do not remove accompaniment, infer permission, or verify that a recording is an isolated vocal.</div>
<details class="panel profile-panel"><summary>Character identity &amp; dataset defaults</summary><form id="profile-form" class="form-grid"><label>Display name<input id="profile-name" required maxlength="80"></label><label>Voice trigger<input id="profile-trigger" required maxlength="64"></label><label>Vocal language<input id="profile-language" placeholder="en, unknown..." required></label><label class="span-two">Default clip caption<input id="profile-caption" required placeholder="Describe the actual vocal sound"></label><div class="align-end"><button class="button secondary requires-character" type="submit">Save identity</button></div><p class="field-help span-all">Defaults apply to the next dataset. Existing clip annotations are kept as-is.</p></form></details>
<div class="dataset-grid">
<article class="panel"><div class="panel-heading"><h2>Source library</h2><span id="source-count" class="pill">0 FILES</span></div><div id="source-list" class="source-list"></div><p id="source-path" class="path-text"></p></article>
<article class="panel prep-panel"><span class="eyebrow">NON-DESTRUCTIVE PREPARATION</span><h2>A little structure.<br>A better starting point.</h2><p class="muted">Silence-aware splitting, 48 kHz stereo for ACE-Step, and 40 kHz mono for SVC.</p><div class="form-grid"><label>Maximum clip length<select id="clip-seconds"><option value="8">8 seconds</option><option value="12" selected>12 seconds</option><option value="20">20 seconds</option><option value="30">30 seconds</option></select></label><label>Silence threshold<select id="silence-db"><option value="-55">-55 dB</option><option value="-45" selected>-45 dB</option><option value="-35">-35 dB</option></select></label></div><button id="prepare-button" class="button primary full requires-character"><svg class="icon"><use href="#i-wave"/></svg>Prepare a new dataset</button><p class="field-help">Creates a new version; it never overwrites an earlier dataset.</p></article>
</div>
<div class="section-heading"><div><h2>Clip review</h2><p class="muted">Select a clip, listen, and annotate what is actually sung.</p></div><span id="review-count" class="pill">NO DATASET YET</span></div>
<div id="dataset-warnings" class="notice warning hidden"></div>
<div class="review-grid">
<article class="panel clip-panel"><div id="clip-list" class="clip-list"></div></article>
<article class="panel annotation-panel"><div class="panel-heading"><h2 id="clip-heading">Choose a clip</h2><span id="clip-duration" class="pill">--</span></div><audio id="clip-audio" controls preload="none"></audio><form id="annotation-form"><label>Caption<textarea id="clip-caption" rows="2" required placeholder="Describe the audible voice and performance."></textarea></label><label>Lyrics / actual vocalisation<textarea id="clip-lyrics" rows="5" placeholder="Accurate syllables or lyrics. For wordless singing, describe the actual vocalisation."></textarea></label><p class="field-help">Never use [Instrumental] for vocal recordings. Unknown lyrics are left blank, not invented.</p><label class="check-label"><input id="clip-reviewed" type="checkbox"><span>I listened to this clip and verified its caption and vocal text.</span></label><div class="button-row"><button id="save-clip" class="button primary" type="submit" disabled>Save clip review</button><button id="next-clip" class="button secondary" type="button" disabled>Next clip<svg class="icon"><use href="#i-arrow"/></svg></button></div></form></article>
</div>
</section>
<section id="page-training" class="page hidden">
<div class="page-heading"><div><p class="eyebrow">A VOICE WITH YOUR SIGNATURE</p><h1>Train your character</h1><p class="muted">An ACE-Step XL adapter for original songs. Your configuration, your checkpoints.</p></div><span class="pill accent">LOCAL GPU WORKFLOW</span></div>
<div class="notice warning"><strong>Two different models.</strong> This trainer creates an ACE-Step adapter. For SVC that preserves an existing performance, export the vocal dataset and train a separate RVC model in Applio.</div>
<div class="training-grid">
<article class="panel"><div class="panel-heading"><h2>Training recipe</h2><span class="pill">XL / 4B</span></div><form id="training-form">
<label>Base model<select id="model-select"><option value="acestep-v15-xl-sft">XL SFT</option><option value="acestep-v15-xl-turbo">XL Turbo</option><option value="acestep-v15-xl-base">XL Base</option></select></label>
<p class="field-help">Keep the adapter paired with the exact XL variant used for training.</p>
<div class="form-grid"><label>Epochs<input id="epochs" type="number" min="1" max="2000" value="100" required></label><label>LoRA rank<select id="rank"><option>8</option><option selected>16</option><option>32</option><option>64</option></select></label><label>Learning rate<input id="learning-rate" type="number" min="0.000001" max="0.001" step="0.000001" value="0.0001" required></label><label>Gradient accumulation<input id="accumulation" type="number" min="1" max="32" value="4" required></label></div>
<div class="recipe-footnote"><svg class="icon"><use href="#i-settings"/></svg><span>Batch 1 &middot; BF16 &middot; Gradient checkpointing<br>Encoder offload &middot; AdamW &middot; Windows-safe workers</span></div>
<label class="check-label authorization"><input id="authorized" type="checkbox"><span>I own or have permission to train and use this voice. I have checked that the clips are suitable isolated vocals.</span></label>
<div class="button-row"><button id="preprocess-button" class="button secondary requires-character" type="button">1. Preprocess clips</button><button id="train-button" class="button primary requires-character" type="submit"><svg class="icon"><use href="#i-spark"/></svg>2. Start training</button></div>
<p class="field-help">These buttons start real backend jobs. All clips require review before preprocessing. No training has run just by opening this page.</p>
</form></article>
<div class="training-aside">
<article class="panel readiness"><div class="panel-heading"><h2>Before you begin</h2><span id="readiness-pill" class="pill">CHECKLIST</span></div><ul id="readiness-list" class="checklist"></ul><p class="field-help">Short clips and batch size 1 are conservative starting settings, not a guarantee that XL training fits in VRAM. Close other GPU workloads before training.</p></article>
<article class="panel command-panel"><div class="panel-heading"><h2>Under the hood</h2><button id="preview-recipe" class="text-button requires-character">Preview command</button></div><pre id="command-preview">Your command preview will appear here.
Previewing does not execute anything.</pre></article>
</div>
</div>
<article class="log-panel"><div class="log-toolbar"><div><span id="job-dot" class="status-dot"></span><strong id="job-title">Training console</strong><span id="job-status" class="pill">IDLE</span></div><select id="job-select" aria-label="Choose a job"><option value="">No jobs yet</option></select></div><pre id="job-log" tabindex="0" aria-label="Job output">Ready when you are.
No jobs have been started for this character.</pre><div class="log-footer"><span>Output comes directly from your local jobs.</span><span id="job-time">No elapsed time to report</span></div></article>
</section>
<section id="page-exports" class="page hidden">
<div class="page-heading"><div><p class="eyebrow">TAKE YOUR CHARACTER INTO THE MUSIC</p><h1>Models &amp; exports</h1><p class="muted">The right artifact for the right workflow. No interchangeable-file assumptions.</p></div></div>
<div class="export-grid">
<article class="panel export-card"><span class="export-icon lavender"><svg class="icon"><use href="#i-spark"/></svg></span><span class="eyebrow">ORIGINAL SONGS</span><h2>ACE-Step XL adapter</h2><p>Use trained weights with the matching XL base model in ACE-Step's browser UI. Include your character's trigger in the generation caption.</p><div class="format-tags"><span>.safetensors</span><span>adapter_config.json</span></div><div class="export-note"><strong id="export-trigger">No character selected</strong><small>Character trigger</small></div><button class="button secondary full" data-page="training">Go to XL training<svg class="icon"><use href="#i-arrow"/></svg></button><p class="field-help">ComfyUI support depends on your loader's XL and adapter-format support. The studio does not silently convert or merge weights.</p></article>
<article class="panel export-card"><span class="export-icon mint"><svg class="icon"><use href="#i-mic"/></svg></span><span class="eyebrow">SINGING VOICE CONVERSION</span><h2>RVC / Applio dataset</h2><p>Preserve a sung performance while changing its vocal identity. Export prepared clips, then train the separate conversion model in Applio.</p><div class="format-tags"><span>40 kHz mono WAV</span><span>.zip dataset</span></div><div class="export-note"><strong>Separate model training required</strong><small>Applio produces .pth + .index files</small></div><button id="export-svc" class="button primary full requires-character"><svg class="icon"><use href="#i-box"/></svg>Build SVC dataset export</button><p class="field-help">This is a dataset export, not a trained SVC model. Use isolated vocals for training and conversion; remix with the instrumental afterwards.</p></article>
</div>
<div class="section-heading"><div><h2>Saved artifacts</h2><p class="muted">Files that actually exist in this character's workspace.</p></div><button id="refresh-artifacts" class="text-button requires-character">Refresh files</button></div>
<div id="artifact-list" class="panel artifact-list"></div>
</section>
<section id="page-settings" class="page hidden">
<div class="page-heading"><div><p class="eyebrow">BUILT AROUND YOUR LOCAL SETUP</p><h1>Workspace settings</h1><p class="muted">Point the studio at your existing installation. No containers or MCP server required.</p></div></div>
<article class="panel settings-panel"><form id="settings-form"><label>ACE-Step installation<input id="settings-ace" required spellcheck="false"></label><label>ACE-Step Python interpreter<input id="settings-python" required spellcheck="false"></label><p class="field-help">Use its prepared environment, not a second conflicting Torch installation.</p><label>Voice character library<input id="settings-source" required spellcheck="false"></label><p class="field-help">Each immediate subfolder is a separate character. Add as many characters as you need.</p><label>Studio output workspace<input id="settings-workspace" readonly></label><p class="field-help">Set VOICE_STUDIO_WORKSPACE before launch to change the output location.</p><button class="button primary" type="submit">Save workspace settings</button></form></article>
<article class="panel setup-notes"><h2>What runs where</h2><div class="setup-grid"><div><h3>Voice Studio</h3><p>A loopback-only browser UI, local dataset tools, and job orchestration. It never starts training automatically.</p></div><div><h3>ACE-Step 1.5</h3><p>The existing XL models and native Side-Step training code. Preprocessing and training run in its Python environment.</p></div><div><h3>Applio / RVC</h3><p>A separate conversion trainer. Import the SVC dataset there; its runtime, pretrained assets, and models stay separate.</p></div></div></article>
</section>
<footer class="main-footer"><span>VOICE STUDIO <span class="footer-dot">&middot;</span> Made for the voice behind the song.</span><span>Local files. Explicit actions. Your control.</span></footer>
</main>
</div>
<dialog id="character-dialog"><form id="character-form"><div class="dialog-heading"><span class="stat-icon lavender"><svg class="icon"><use href="#i-mic"/></svg></span><button class="icon-button" type="button" id="close-dialog" aria-label="Close dialog">&times;</button></div><p class="eyebrow">EVERY VOICE HAS A STORY</p><h2>Meet your next character.</h2><p class="muted">Give this voice its own place in your library. Add recordings when you're ready.</p><label>Character name<input id="new-name" required maxlength="80" placeholder="e.g. Ember, Midnight Alto, Folk Tenor"></label><label>Voice trigger <span class="optional">optional</span><input id="new-trigger" maxlength="64" placeholder="e.g. ember_voice"></label><p class="field-help">A unique tag to use in ACE-Step generation captions. Letters, numbers, hyphens, and underscores.</p><p id="dialog-error" class="form-error" role="alert"></p><button type="submit" class="button primary full"><svg class="icon"><use href="#i-plus"/></svg>Create character</button></form></dialog>
<div id="toast" class="toast hidden" role="status" aria-live="polite"></div>
</body>
</html>

View File

@ -0,0 +1,254 @@
:root {
--bg: #f5f4f9; --paper: #fff; --ink: #252238; --muted: #7c788e;
--purple: #7755d8; --purple-dark: #5d3cbb; --line: #eae7f1;
--lavender: #eee8fd; --mint: #e6f3ee; --green: #36876c;
--radius: 16px; --shadow: 0 5px 24px #30224b04;
font-family: "Segoe UI", -apple-system, BlinkMacSystemFont, sans-serif;
color: var(--ink); background: var(--bg); font-size: 14px;
}
* { box-sizing: border-box; }
body { margin: 0; }
button, input, select, textarea { font: inherit; }
button, a, summary, select { -webkit-tap-highlight-color: transparent; }
button { cursor: pointer; }
button:disabled { opacity: .45; cursor: not-allowed; }
button, a, input, textarea, select, summary { outline-offset: 4px; }
:focus-visible { outline: 2px solid var(--purple); }
button { transition: background .16s, transform .16s, box-shadow .16s; }
button:active:not(:disabled) { transform: translateY(1px); }
a { color: var(--purple); }
h1, h2, h3, p { margin: 0; }
h1 { font-size: clamp(25px, 2.2vw, 34px); letter-spacing: -1.1px; font-weight: 650; margin-bottom: 10px; }
h2 { font-size: 19px; letter-spacing: -.45px; font-weight: 650; }
h3 { font-size: 15px; font-weight: 650; }
p { line-height: 1.75; }
small { color: var(--muted); }
.hidden { display: none !important; }
.icon-library { position: absolute; width: 0; height: 0; overflow: hidden; }
.icon { width: 19px; height: 19px; fill: none; stroke: currentColor; stroke-width: 1.65; stroke-linecap: round; stroke-linejoin: round; flex-shrink: 0; }
.sidebar { width: 252px; background: #fbfafe; border-right: 1px solid var(--line); position: fixed; inset: 0 auto 0 0; padding: 34px 18px 20px; display: flex; flex-direction: column; z-index: 20; }
.brand { display: flex; align-items: center; gap: 11px; text-decoration: none; color: var(--ink); font-size: 24px; font-weight: 750; letter-spacing: -1px; margin: 0 12px 43px; }
.brand-light { font-weight: 400; }
.brand small { display: block; font-size: 8px; letter-spacing: 2.3px; margin-top: 5px; font-weight: 600; }
.brand-mark { display: grid; place-items: center; width: 39px; height: 43px; background: linear-gradient(145deg,#9977ed,#7151c5); color: white; border-radius: 12px; box-shadow: 0 5px 12px #7755d830; }
.brand-mark .icon { width: 25px; height: 25px; }
.workspace-label, .library-label { color: #9691a8; font-size: 10px; letter-spacing: 1.5px; font-weight: 650; padding: 0 14px; }
.workspace-label { display: flex; align-items: center; justify-content: space-between; margin-bottom: 17px; }
.local-dot { color: #888095; font-size: 8px; letter-spacing: .8px; display: flex; align-items: center; gap: 5px; }
.local-dot::before { content: ""; width: 5px; height: 5px; background: #6aa58a; border-radius: 50%; }
.navigation { display: flex; flex-direction: column; gap: 5px; }
.nav-item { border: 0; background: transparent; width: 100%; display: flex; align-items: center; gap: 12px; color: #767087; padding: 13px 14px; text-align: left; border-radius: 9px; font-size: 12px; font-weight: 550; }
.nav-item:hover { background: #f0ecf7; }
.nav-item.active { background: #eee7fb; color: var(--purple); }
.nav-count { background: #fff9; font-size: 10px; padding: 2px 6px; border-radius: 5px; margin-left: auto; }
.library-label { display: flex; align-items: center; justify-content: space-between; margin: 34px 0 9px; padding-right: 5px; }
.icon-button { border: 0; background: transparent; color: var(--muted); padding: 6px; border-radius: 7px; display: inline-grid; place-items: center; font-size: 24px; }
.icon-button:hover { color: var(--purple); background: var(--lavender); }
.character-list { overflow-y: auto; min-height: 40px; max-height: 28vh; }
.character-item { display: flex; align-items: center; gap: 10px; border: none; background: none; padding: 10px 13px; border-radius: 9px; text-align: left; width: 100%; color: #777184; }
.character-item:hover, .character-item.selected { background: #f1edf8; color: var(--ink); }
.avatar { width: 34px; height: 34px; display: grid; place-items: center; background: #e9e2f7; color: #7d62ab; border-radius: 10px; font-weight: 650; font-size: 12px; flex-shrink: 0; }
.character-item:nth-child(3n+2) .avatar { background: #e4eee9; color: #5c8875; }
.character-item:nth-child(3n) .avatar { background: #f6e9e3; color: #b88d77; }
.character-text { overflow: hidden; }
.character-text strong { display: block; font-size: 11px; font-weight: 600; white-space: nowrap; text-overflow: ellipsis; overflow: hidden; }
.character-text small { display: block; font-size: 10px; margin-top: 4px; }
.sidebar-add { margin: 10px 12px 0; display: flex; align-items: center; gap: 9px; border: none; background: none; font-size: 11px; color: #9488ab; padding: 8px 0; }
.sidebar-add .icon { width: 15px; height: 15px; }
.sidebar-bottom { margin-top: auto; padding-top: 20px; }
.privacy-card { background: #f0ecf7; border: 1px solid #e8e1f2; border-radius: 12px; padding: 14px 10px; margin: 0 4px 18px; display: flex; align-items: flex-start; gap: 8px; }
.privacy-icon { color: #9983b9; padding-top: 2px; }
.privacy-card strong { font-size: 10px; color: #6b5c82; font-weight: 600; }
.privacy-card p { font-size: 9px; color: #9b8dab; margin-top: 3px; }
.sidebar-version { padding: 21px 14px 0; display: flex; justify-content: space-between; font-size: 8px; letter-spacing: 1px; color: #aaa4b9; }
.main-shell { margin-left: 252px; }
.topbar { height: 77px; display: flex; align-items: center; justify-content: space-between; padding: 0 42px; border-bottom: 1px solid var(--line); background: #ffffffa8; }
.breadcrumb { display: flex; align-items: center; gap: 13px; font-size: 11px; color: #a19aae; }
.breadcrumb strong { color: #6d647e; font-weight: 500; }
.breadcrumb span { color: #c3bace; }
.topbar-right { display: flex; align-items: center; gap: 22px; }
.connection { font-size: 10px; color: #92899e; display: flex; align-items: center; gap: 7px; }
.connection i { width: 6px; height: 6px; border-radius: 50%; background: #aaa2b8; }
.connection.ready i { background: #5fa486; }
.avatar.mini { width: 31px; height: 31px; font-size: 10px; background: #f0eafa; }
main { padding: 36px 42px 0; max-width: 1550px; margin: auto; }
.page-heading { display: flex; justify-content: space-between; gap: 20px; align-items: center; margin: 3px 0 28px; }
.eyebrow { font-size: 9px; letter-spacing: 2px; color: #9b8cb4; font-weight: 650; margin-bottom: 12px; }
.muted { color: var(--muted); font-size: 12px; }
.button { border: 1px solid transparent; padding: 11px 16px; display: inline-flex; align-items: center; justify-content: center; gap: 9px; border-radius: 9px; font-size: 11px; font-weight: 600; text-decoration: none; white-space: nowrap; line-height: 1.5; }
.button .icon { width: 16px; height: 16px; }
.button.primary { background: var(--purple); color: #fff; box-shadow: 0 4px 11px #7755d81c; }
.button.primary:hover { background: var(--purple-dark); }
.button.secondary { background: white; color: #716584; border-color: #e4ddeb; }
.button.secondary:hover { background: #f8f5fd; }
.button.white { background: #fff; color: #7959bc; }
.button.ghost-light { background: #ffffff0d; color: #eee6ff; border-color: #ffffff24; }
.full { width: 100%; }
.text-button { display: inline-flex; align-items: center; gap: 8px; font-size: 10px; font-weight: 600; color: var(--purple); border: none; background: transparent; padding: 5px 0; }
.text-button .icon { width: 14px; height: 14px; }
.hero { display: flex; justify-content: space-between; align-items: center; min-height: 268px; padding: 33px 38px; border-radius: 18px; background: linear-gradient(112deg,#8061bf,#9672cc 57%,#c1a3df); position: relative; overflow: hidden; box-shadow: 0 12px 26px #71519215; }
.hero::after { content: ""; position: absolute; width: 500px; height: 500px; border-radius: 50%; border: 1px solid #ffffff0b; right: -40px; top: -130px; pointer-events: none; }
.hero-content { z-index: 1; }
.hero h2 { color: #fff; font-size: clamp(24px,2.3vw,34px); letter-spacing: -1px; margin: 18px 0 10px; overflow-wrap: anywhere; }
.hero p { color: #eee5fc; font-size: 12px; line-height: 1.9; }
.hero-actions { display: flex; gap: 12px; margin-top: 23px; flex-wrap: wrap; }
.pill { border: 1px solid var(--line); background: #faf9fc; color: #9690a4; padding: 5px 8px; border-radius: 6px; font-size: 8px; font-weight: 600; letter-spacing: .7px; white-space: nowrap; display: inline-block; line-height: 1.3; }
.hero-pill { background: #ffffff15; color: #f1eafa; border-color: #ffffff25; font-size: 8px; letter-spacing: 1.5px; }
.pill.accent { background: var(--lavender); border-color: #e6ddfa; color: var(--purple); }
.pill.success { background: var(--mint); border-color: #d7e9e1; color: var(--green); }
.pill.failed { background: #fbe8e8; color: #b65353; border-color: #f2d5d5; }
.voice-art { width: 250px; height: 210px; position: relative; flex-shrink: 0; display: grid; place-items: center; margin-right: 20px; }
.orbit { position: absolute; border: 1px solid #ffffff25; border-radius: 50%; }
.orbit-one { width: 205px; height: 205px; }
.orbit-two { width: 170px; height: 170px; border-color: #ffffff35; }
.orbit-three { width: 135px; height: 135px; border-color: #ffffff50; }
.voice-core { width: 106px; height: 106px; border-radius: 50%; display: grid; place-items: center; background: radial-gradient(circle at 30% 20%,#e2d1f277,#ffffff12); border: 1px solid #fff6; box-shadow: 0 0 70px #fff2,inset 0 0 18px #ffffff18; backdrop-filter: blur(10px); }
.voice-core svg { width: 65px; height: 65px; fill: none; stroke: #fff; stroke-width: 1.2; stroke-linecap: round; }
.art-label { position: absolute; bottom: -9px; font-size: 7px; letter-spacing: 2px; color: #f0dffa; }
.stat-grid { display: grid; grid-template-columns: repeat(4,1fr); gap: 16px; margin: 22px 0 34px; }
.stat { display: flex; gap: 13px; background: white; border: 1px solid var(--line); border-radius: 12px; padding: 19px 16px; box-shadow: var(--shadow); align-items: flex-start; }
.stat-icon, .export-icon { width: 36px; height: 36px; border-radius: 10px; display: grid; place-items: center; flex-shrink: 0; }
.lavender { background: var(--lavender); color: #9573d1; }
.peach { background: #fbede5; color: #dca57d; }
.mint { background: var(--mint); color: #77a992; }
.blue { background: #e9effb; color: #86a0cf; }
.stat-label { font-size: 9px; color: #9b94a8; display: block; padding-top: 1px; }
.stat strong { display: block; font-size: 23px; font-weight: 600; letter-spacing: -.7px; margin: 6px 0 5px; }
.stat small { font-size: 8px; display: block; }
.section-heading { display: flex; justify-content: space-between; align-items: center; margin: 30px 0 17px; gap: 20px; }
.section-heading h2 { font-size: 16px; margin-bottom: 5px; }
.section-heading p { font-size: 10px; }
.workflow-grid { display: grid; grid-template-columns: repeat(3,1fr); gap: 18px; }
.step-card { background: #fff; padding: 22px; border: 1px solid var(--line); border-radius: 13px; position: relative; }
.step-number { position: absolute; top: 20px; right: 20px; color: #e5dff0; font-size: 25px; font-weight: 600; }
.step-card h3 { margin-top: 22px; margin-bottom: 10px; }
.step-card p { font-size: 11px; color: #8e879c; margin-bottom: 18px; line-height: 1.85; }
.library-grid { display: grid; grid-template-columns: repeat(3,1fr); gap: 16px; }
.library-card { border: 1px solid var(--line); background: #fff; border-radius: 13px; padding: 20px; display: flex; align-items: center; gap: 13px; text-align: left; color: var(--ink); min-width: 0; }
.library-card:hover { border-color: #cbb8eb; box-shadow: 0 5px 20px #7755d80b; }
.library-card .avatar { width: 45px; height: 45px; border-radius: 12px; }
.library-card h3 { font-size: 12px; overflow-wrap: anywhere; }
.library-card p { font-size: 10px; margin-top: 4px; color: #a197ad; }
.library-card .icon { margin-left: auto; color: #b7aac9; width: 16px; }
.panel { background: var(--paper); border: 1px solid var(--line); border-radius: var(--radius); padding: 25px; box-shadow: var(--shadow); }
.panel-heading { display: flex; justify-content: space-between; align-items: center; gap: 14px; margin-bottom: 22px; }
.panel-heading h2 { font-size: 16px; }
.notice { padding: 14px 18px; border: 1px solid #e5ddf3; background: #f0ebf8; color: #837098; font-size: 11px; line-height: 1.8; border-radius: 10px; margin-bottom: 22px; overflow-wrap: anywhere; }
.notice strong { color: #67547f; font-weight: 650; }
.notice.warning { background: #fcf5e9; color: #9b835b; border-color: #f0e5d0; }
.notice.warning strong { color: #8b724b; }
.notice.danger { background: #fceeee; border-color: #f0d7d7; color: #af5353; }
.dataset-grid { display: grid; grid-template-columns: 1.4fr 1fr; gap: 22px; margin-top: 20px; }
.prep-panel { background: linear-gradient(135deg,#fff,#f8f4fe); }
.prep-panel h2 { font-size: 23px; margin-bottom: 15px; line-height: 1.4; }
.prep-panel .form-grid { margin: 24px 0 8px; }
.prep-panel .field-help { margin-top: 10px; text-align: center; }
.form-grid { display: grid; grid-template-columns: 1fr 1fr; gap: 18px; }
label { display: flex; flex-direction: column; gap: 8px; font-size: 10px; font-weight: 600; color: #766a89; margin-bottom: 18px; }
input, select, textarea { color: #635870; background: #fcfbfe; border: 1px solid #e7e1ee; border-radius: 8px; padding: 10px 12px; width: 100%; font-size: 12px; min-width: 0; }
textarea { resize: vertical; line-height: 1.7; }
input:focus, select:focus, textarea:focus { border-color: #bda5e2; background: white; }
input[readonly] { color: #a79bb5; background: #f7f5fa; }
.form-grid label { margin: 0; }
.field-help { color: #a296ae; font-size: 10px; line-height: 1.8; margin: -8px 0 18px; font-weight: 400; }
.span-two { grid-column: span 2; }
.span-all { grid-column: 1 / -1; }
.align-end { display: flex; align-items: end; }
.profile-panel summary { color: #7f6a96; cursor: pointer; font-size: 12px; font-weight: 600; }
.profile-panel[open] summary { margin-bottom: 24px; }
.profile-panel { padding: 20px 25px; }
.source-list { max-height: 290px; overflow-y: auto; }
.source-row { display: flex; gap: 12px; align-items: center; padding: 13px 0; border-bottom: 1px solid #f0ecf5; }
.source-row:last-child { border: 0; }
.source-row .source-details { min-width: 0; flex: 1; }
.source-row strong { font-size: 10px; display: block; overflow-wrap: anywhere; font-weight: 600; }
.source-row small { font-size: 9px; margin-top: 5px; display: block; color: #a69bb2; }
.source-row audio { height: 28px; width: 170px; max-width: 43%; }
.source-row .file-icon { background: #f2edf9; color: #b19bcd; width: 30px; height: 34px; border-radius: 6px; display: grid; place-items: center; }
.source-row .icon { width: 16px; height: 16px; }
.path-text { font: 9px/1.8 Consolas, monospace; color: #aaa0b6; overflow-wrap: anywhere; margin-top: 16px; }
.review-grid { display: grid; grid-template-columns: .8fr 1.4fr; gap: 22px; }
.clip-panel { padding: 10px; }
.clip-list { max-height: 540px; overflow-y: auto; }
.clip-item { width: 100%; padding: 13px 14px; border: none; background: none; border-radius: 9px; color: #8a7f97; display: flex; align-items: center; gap: 12px; text-align: left; }
.clip-item.selected { background: #f0eafa; color: var(--purple); }
.clip-item:hover { background: #f7f3fc; }
.clip-item strong { font-size: 11px; font-weight: 600; }
.clip-item small { margin-left: auto; font-size: 9px; }
.clip-item .icon { width: 15px; height: 15px; color: #b2a1c6; }
.clip-item.reviewed .icon { color: #6ea58b; }
.annotation-panel audio { width: 100%; margin: 0 0 25px; height: 40px; }
.check-label { flex-direction: row; align-items: flex-start; font-size: 11px; font-weight: 400; line-height: 1.8; gap: 10px; }
.check-label input { width: 15px; height: 15px; margin: 3px 0 0; accent-color: var(--purple); flex-shrink: 0; }
.button-row { display: flex; gap: 10px; flex-wrap: wrap; margin-top: 18px; }
.training-grid { display: grid; grid-template-columns: 1.1fr 1fr; gap: 22px; }
.training-aside { display: flex; flex-direction: column; gap: 22px; }
.training-grid .form-grid { margin: 20px 0; }
.recipe-footnote { display: flex; align-items: center; gap: 12px; padding: 15px; background: #f7f4fb; color: #a08bb7; border-radius: 9px; font-size: 10px; line-height: 1.9; margin-bottom: 22px; }
.authorization { border-top: 1px solid var(--line); padding-top: 20px; margin-bottom: 20px; }
#training-form > .field-help:last-child { margin: 15px 0 0; }
.checklist { list-style: none; padding: 0; margin: 0 0 24px; display: flex; flex-direction: column; gap: 18px; }
.checklist li { display: flex; align-items: center; gap: 10px; color: #a196ad; font-size: 11px; }
.checklist .check-circle { width: 20px; height: 20px; border-radius: 50%; display: grid; place-items: center; background: #f1edf6; color: #b7a9c7; font-size: 9px; }
.checklist li.complete { color: #728b7e; }
.checklist li.complete .check-circle { background: #e9f4ed; color: #6d9f82; }
.check-circle .icon { width: 12px; height: 12px; }
.readiness .field-help { margin: 0; }
.command-panel { flex: 1; }
.command-panel pre { font: 10px/1.9 Consolas, monospace; color: #9b8aad; padding: 16px; border-radius: 8px; border: 1px dashed #e5dced; background: #fbf9fe; white-space: pre-wrap; overflow-wrap: anywhere; max-height: 190px; overflow-y: auto; }
.log-panel { border: 1px solid #332e42; border-radius: 13px; background: #24202f; margin-top: 25px; overflow: hidden; }
.log-toolbar { background: #2d273a; padding: 16px 20px; display: flex; align-items: center; justify-content: space-between; gap: 15px; }
.log-toolbar > div { display: flex; align-items: center; gap: 10px; }
.log-toolbar strong { font-size: 11px; color: #c1b5d2; }
.log-toolbar .pill { background: #ffffff08; color: #a999bd; border-color: #ffffff12; font-size: 7px; }
.log-toolbar select { width: 210px; background: #342d43; border-color: #494056; color: #b4a5c6; font-size: 9px; padding: 7px 9px; }
.status-dot { width: 6px; height: 6px; border-radius: 50%; background: #81748f; display: inline-block; }
.status-dot.running { background: #af91ec; box-shadow: 0 0 0 4px #af91ec16; }
.status-dot.completed { background: #88b79a; }
.status-dot.failed { background: #dd8c8c; }
#job-log { font: 11px/1.9 Consolas, monospace; color: #b6a8c7; padding: 20px; margin: 0; min-height: 190px; max-height: 360px; overflow: auto; white-space: pre-wrap; overflow-wrap: anywhere; }
.log-footer { padding: 12px 20px; border-top: 1px solid #ffffff07; display: flex; justify-content: space-between; gap: 15px; color: #776987; font-size: 9px; }
.export-grid { display: grid; grid-template-columns: 1fr 1fr; gap: 22px; }
.export-card { position: relative; padding: 30px; }
.export-icon { width: 48px; height: 48px; margin-bottom: 24px; border-radius: 13px; }
.export-icon .icon { width: 24px; height: 24px; }
.export-card h2 { font-size: 24px; margin-bottom: 15px; }
.export-card > p:not(.field-help) { color: #94879f; font-size: 12px; line-height: 1.9; }
.format-tags { display: flex; gap: 8px; margin: 22px 0; flex-wrap: wrap; }
.format-tags span { font: 9px Consolas, monospace; color: #a48db9; background: #f7f3fb; border: 1px solid #eee7f4; padding: 7px 10px; border-radius: 6px; }
.export-note { padding: 16px; border: 1px solid #ece5f3; border-radius: 9px; margin-bottom: 23px; background: #fcfafe; }
.export-note strong { font-size: 11px; font-weight: 600; color: #8d76a3; overflow-wrap: anywhere; }
.export-note small { display: block; margin-top: 5px; font-size: 9px; }
.export-card .field-help { margin: 17px 0 0; }
.artifact-list { padding: 8px 25px; }
.artifact-row { display: flex; gap: 15px; align-items: center; border-bottom: 1px solid var(--line); padding: 17px 0; }
.artifact-row:last-child { border-bottom: none; }
.artifact-detail { flex: 1; min-width: 0; }
.artifact-detail strong { display: block; font-size: 11px; font-weight: 600; }
.artifact-detail small { display: block; margin-top: 5px; font-size: 9px; overflow-wrap: anywhere; }
.artifact-size { font-size: 10px; color: #b4a7bf; white-space: nowrap; }
.settings-panel { max-width: 800px; }
.settings-panel input { font-family: Consolas, monospace; }
.setup-notes { margin-top: 25px; }
.setup-grid { display: grid; grid-template-columns: repeat(3,1fr); gap: 30px; margin-top: 24px; }
.setup-grid h3 { font-size: 12px; margin-bottom: 10px; }
.setup-grid p { font-size: 11px; color: #a396ac; line-height: 1.9; }
.empty-state { padding: 35px 20px; text-align: center; color: #b4a7bf; font-size: 11px; line-height: 1.9; width: 100%; }
.library-grid > .empty-state { grid-column: 1 / -1; border: 1px dashed #dfd5ea; border-radius: 12px; }
.main-footer { display: flex; justify-content: space-between; gap: 20px; padding: 28px 0 22px; margin-top: 30px; border-top: 1px solid var(--line); font-size: 8px; color: #b8adc4; }
.main-footer > span:first-child { letter-spacing: .4px; }
.footer-dot { padding: 0 8px; }
dialog { border: 1px solid #e9e0f3; border-radius: 20px; width: min(480px,calc(100vw - 32px)); padding: 32px; color: var(--ink); box-shadow: 0 30px 100px #26173730; }
dialog::backdrop { background: #25193260; backdrop-filter: blur(5px); }
.dialog-heading { display: flex; justify-content: space-between; align-items: center; margin-bottom: 24px; }
dialog h2 { font-size: 26px; letter-spacing: -.8px; }
dialog .muted { margin: 15px 0 24px; }
.optional { font-size: 9px; font-weight: 400; color: #b5a7c3; }
.form-error { font-size: 11px; color: #b95b62; margin-bottom: 15px; }
.toast { position: fixed; bottom: 25px; right: 30px; background: #fff; border: 1px solid #ded3ec; color: #78618e; box-shadow: 0 10px 45px #2a153b1a; padding: 15px 22px; border-radius: 12px; z-index: 100; font-size: 12px; max-width: min(500px,calc(100vw - 40px)); line-height: 1.7; }
.toast.error { color: #ad525d; border-color: #efcdd1; }
@media (min-width: 1600px) { main { padding-top: 44px; } .hero { min-height: 295px; } .stat { padding: 24px 20px; } }
@media (max-width: 1200px) { .sidebar { width: 220px; padding-left: 12px; padding-right: 12px; } .main-shell { margin-left: 220px; } main { padding: 28px 25px 0; } .topbar { padding: 0 25px; } .stat-grid { grid-template-columns: repeat(2,1fr); } .hero { padding: 30px; } .voice-art { width: 190px; margin-right: 0; } .workflow-grid { gap: 12px; } .step-card { padding: 18px; } .library-grid { grid-template-columns: repeat(2,1fr); } .dataset-grid,.training-grid { grid-template-columns: 1fr; } .source-list { max-height: 250px; } }
@media (max-width: 820px) { .sidebar { width: 78px; padding: 25px 10px; } .brand { margin: 0 7px 35px; } .brand > span:last-child,.workspace-label,.library-label,.character-list,.sidebar-add,.privacy-card,.sidebar-version,.nav-count { display: none; } .nav-item { font-size: 0; justify-content: center; padding: 14px; } .nav-item .icon { width: 21px; height: 21px; } .main-shell { margin-left: 78px; } .hero h2 { font-size: 28px; } .voice-art { display: none; } .workflow-grid { grid-template-columns: 1fr; } .step-card p { max-width: 90%; } .step-card h3 { margin-top: 18px; } .export-grid,.review-grid { grid-template-columns: 1fr; } .clip-list { max-height: 230px; } .setup-grid { grid-template-columns: 1fr; gap: 20px; } .page-heading { align-items: flex-start; } .page-heading > .pill { display: none; } .log-toolbar { flex-wrap: wrap; } .log-toolbar select { width: 100%; } .topbar { height: 65px; } .main-footer > span:last-child { display: none; } }
@media (max-width: 540px) { main { padding: 23px 16px 0; } .topbar { padding: 0 16px; } .connection { font-size: 0; gap: 0; } .topbar-right { gap: 12px; } .breadcrumb { font-size: 9px; gap: 8px; } .page-heading { flex-wrap: wrap; gap: 12px; } h1 { font-size: 25px; } .hero { padding: 27px 22px; } .hero h2 { font-size: 25px; } .hero p { font-size: 11px; } .hero-actions { gap: 8px; } .hero-actions .button { font-size: 10px; padding: 10px; } .stat-grid { gap: 10px; } .stat { flex-direction: column; padding: 15px; gap: 10px; } .stat strong { font-size: 22px; } .stat-label { font-size: 9px; } .library-grid { grid-template-columns: 1fr; } .panel { padding: 20px; } .clip-panel { padding: 8px; } .form-grid { grid-template-columns: 1fr; } .span-two { grid-column: auto; } .source-row { flex-wrap: wrap; } .source-row audio { width: 100%; max-width: 100%; } .source-row .source-details { flex-basis: 70%; } .section-heading { align-items: flex-start; } .artifact-row { flex-wrap: wrap; gap: 10px; } .artifact-size { display: none; } .artifact-detail { flex-basis: 80%; } .log-footer { flex-direction: column; } }
@media (prefers-reduced-motion: reduce) { * { scroll-behavior: auto !important; transition: none !important; } }

View File

@ -0,0 +1,549 @@
from __future__ import annotations
import hashlib
import json
import math
import os
import re
import subprocess
import threading
import traceback
import uuid
import zipfile
from datetime import datetime, timezone
from pathlib import Path
from typing import Callable
import numpy as np
import soundfile as sf
from scipy.signal import resample_poly
ROOT = Path(__file__).resolve().parents[2]
WORKSPACE = Path(os.environ.get("VOICE_STUDIO_WORKSPACE", str(ROOT / "workspace"))).resolve()
AUDIO_SUFFIXES = {".wav", ".flac", ".ogg"}
MODELS = {
"acestep-v15-xl-sft": "xl_sft",
"acestep-v15-xl-turbo": "xl_turbo",
"acestep-v15-xl-base": "xl_base",
}
def now() -> str:
return datetime.now(timezone.utc).isoformat(timespec="seconds")
def write_json(path: Path, value: object) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
temporary = path.with_name(f".{path.name}.{uuid.uuid4().hex}.tmp")
temporary.write_text(json.dumps(value, indent=2, ensure_ascii=False), encoding="utf-8")
temporary.replace(path)
def read_json(path: Path) -> dict:
return json.loads(path.read_text(encoding="utf-8"))
def fingerprint(path: Path) -> str:
return hashlib.sha256(path.read_bytes()).hexdigest()
def identifier(path: Path) -> str:
return hashlib.sha256(os.path.normcase(str(path.resolve())).encode()).hexdigest()[:16]
def slug(value: str) -> str:
result = re.sub(r"[^a-zA-Z0-9_-]+", "-", value.strip()).strip("-")[:64]
if not result or result in {".", ".."}:
raise ValueError("Use a name containing letters or numbers.")
if result.upper().split(".")[0] in {
"CON", "PRN", "AUX", "NUL", *(f"COM{i}" for i in range(1, 10)),
*(f"LPT{i}" for i in range(1, 10)),
}:
raise ValueError("This name is reserved by Windows.")
return result
def settings() -> dict:
saved = read_json(WORKSPACE / "settings.json") if (WORKSPACE / "settings.json").is_file() else {}
ace = Path(saved.get("ace_root", os.environ.get("ACESTEP_ROOT", r"D:\ai\ACE-Step-1.5"))).resolve()
return {
"ace_root": str(ace),
"ace_python": saved.get("ace_python", str(ace / ".venv" / "Scripts" / "python.exe")),
"source_root": saved.get(
"source_root", os.environ.get("VOICE_STUDIO_SOURCE", str(ROOT.parent / "Voice Characters"))
),
"workspace": str(WORKSPACE),
}
def inventory(source: Path) -> list[dict]:
recordings = []
for path in sorted(source.rglob("*")):
if not path.is_file() or path.suffix.lower() not in AUDIO_SUFFIXES:
continue
entry = {
"id": identifier(path), "name": str(path.relative_to(source)),
"bytes": path.stat().st_size, "path": str(path),
}
try:
info = sf.info(str(path))
entry.update(duration=round(info.duration, 3), sample_rate=info.samplerate,
channels=info.channels, subtype=info.subtype, error=None)
except (RuntimeError, sf.LibsndfileError) as exc:
entry.update(duration=0, sample_rate=0, channels=0, subtype="", error=str(exc))
recordings.append(entry)
return recordings
def characters() -> list[dict]:
source_root = Path(settings()["source_root"])
if not source_root.is_dir():
return []
result = []
for source in sorted(source_root.iterdir()):
if not source.is_dir() or source.name.startswith("."):
continue
cid = identifier(source)
config_path = WORKSPACE / cid / "character.json"
config = read_json(config_path) if config_path.is_file() else {}
files = inventory(source)
result.append({
"id": cid, "name": config.get("name", source.name), "source": str(source.resolve()),
"trigger": config.get("trigger") or re.sub(r"[^a-zA-Z0-9_-]+", "-", source.name).strip("-")[:64] or f"voice-{cid}",
"caption": config.get("caption", "Solo sung vocal performance."),
"language": config.get("language", "unknown"), "recordings": files,
"duration": round(sum(item["duration"] for item in files), 2),
"dataset": config.get("dataset"), "created_at": config.get("created_at"),
})
return result
def character(cid: str) -> dict:
for item in characters():
if item["id"] == cid:
return item
raise FileNotFoundError("Voice character not found. Refresh the character library.")
def save_character(item: dict) -> None:
stored = {key: item.get(key) for key in (
"name", "trigger", "caption", "language", "dataset", "created_at"
)}
write_json(WORKSPACE / item["id"] / "character.json", stored)
def create_character(name: str, trigger: str) -> dict:
folder_name = slug(name)
tag = slug(trigger or folder_name)
source = Path(settings()["source_root"]) / folder_name
source.mkdir(parents=True, exist_ok=False)
item = character(identifier(source))
item.update(name=name.strip(), trigger=tag, created_at=now())
save_character(item)
return item
def dataset_path(item: dict) -> Path:
if not item.get("dataset"):
raise ValueError("Prepare a dataset for this character first.")
path = (WORKSPACE / item["id"] / item["dataset"]).resolve()
if not path.is_relative_to(WORKSPACE / item["id"]) or not path.is_file():
raise ValueError("The prepared dataset is missing or outside this character's workspace.")
return path
def dataset_view(item: dict) -> dict:
if not item.get("dataset"):
return {"samples": [], "ready": False, "reviewed": 0, "duration": 0}
path = dataset_path(item)
data = read_json(path)
ready_file = path.parent / "ready.json"
ready = read_json(ready_file) if ready_file.is_file() else {}
return {
**data, "path": str(path), "version": path.parent.name,
"reviewed": sum(bool(s.get("reviewed")) for s in data["samples"]),
"duration": round(sum(s["duration"] for s in data["samples"]), 2),
"ready": ready if ready.get("dataset_sha256") == fingerprint(path) else False,
}
def segment_ranges(audio: np.ndarray, rate: int, max_seconds: float,
silence_db: float) -> list[tuple[int, int]]:
window = max(1, round(rate * 0.02))
count = math.ceil(len(audio) / window)
energy = np.array([
np.sqrt(np.mean(np.square(audio[i * window:(i + 1) * window], dtype=np.float64)))
for i in range(count)
])
active = np.flatnonzero(energy > 10 ** (silence_db / 20))
if not len(active):
return []
groups: list[tuple[int, int]] = []
start = previous = int(active[0])
for index in active[1:]:
index = int(index)
if index - previous > 18:
groups.append((start, previous + 1))
start = index
previous = index
groups.append((start, previous + 1))
padded: list[tuple[int, int]] = []
for start, end in groups:
begin, finish = max(0, (start - 6) * window), min(len(audio), (end + 6) * window)
if padded and begin <= padded[-1][1]:
padded[-1] = (padded[-1][0], finish)
else:
padded.append((begin, finish))
maximum = max(window, int(max_seconds * rate))
minimum = int(1.5 * rate)
result: list[tuple[int, int]] = []
for begin, finish in padded:
while finish - begin > maximum:
target = begin + maximum
lo = max(begin + minimum, target - round(rate))
candidates = energy[math.ceil(lo / window):target // window]
cut = ((math.ceil(lo / window) + int(np.argmin(candidates))) * window
if len(candidates) else target)
if finish - cut < minimum:
cut = begin + (finish - begin) // 2
result.append((begin, cut))
begin = cut
if finish - begin >= minimum:
result.append((begin, finish))
return result
def converted_audio(audio: np.ndarray, source_rate: int, target_rate: int) -> np.ndarray:
if source_rate == target_rate:
result = audio
else:
common = math.gcd(source_rate, target_rate)
result = resample_poly(audio, target_rate // common, source_rate // common, axis=0)
peak = float(np.max(np.abs(result))) if len(result) else 0
return result / peak * 0.999 if peak >= 1 else result
def prepare_dataset(item: dict, clip_seconds: float, silence_db: float,
log: Callable[[str], None]) -> Path:
files = item["recordings"]
if not files:
raise ValueError("Add WAV, FLAC, or OGG recordings to this character first.")
broken = [entry["name"] for entry in files if entry["error"]]
if broken:
raise ValueError(f"Unreadable recordings: {', '.join(broken)}")
run_id = datetime.now().strftime("%Y%m%d-%H%M%S") + "-" + uuid.uuid4().hex[:8]
output = WORKSPACE / item["id"] / "datasets" / run_id
ace_dir, svc_dir = output / "ace_audio", output / "svc_audio"
ace_dir.mkdir(parents=True)
svc_dir.mkdir()
samples, provenance, warnings = [], [], []
for entry in files:
log(f"Reading {entry['name']}")
audio, rate = sf.read(entry["path"], dtype="float32", always_2d=True)
if not len(audio) or audio.shape[1] > 2 or not np.isfinite(audio).all():
raise ValueError(f"{entry['name']}: expected finite mono/stereo audio.")
clipping = float(np.mean(np.abs(audio) >= 0.999))
if clipping > 0.001:
warnings.append(f"{entry['name']}: {clipping:.1%} of samples near full scale; review clipping.")
ranges = segment_ranges(audio, rate, clip_seconds, silence_db)
if not ranges:
warnings.append(f"{entry['name']}: no vocal-length regions above threshold.")
for start, end in ranges:
clip = audio[start:end]
mono = clip.mean(axis=1)
channel_energy = np.mean(np.square(clip, dtype=np.float64), axis=0)
if float(np.mean(np.square(mono, dtype=np.float64))) < float(channel_energy.max()) * 0.25:
mono = clip[:, int(np.argmax(channel_energy))]
warnings.append(f"{entry['name']}: SVC downmix used the strongest channel to avoid cancellation.")
name = f"{len(samples) + 1:05d}.wav"
stereo = np.repeat(clip, 2, axis=1) if clip.shape[1] == 1 else clip
sf.write(str(ace_dir / name), converted_audio(stereo, rate, 48000), 48000, subtype="PCM_16")
sf.write(str(svc_dir / name), converted_audio(mono, rate, 40000), 40000, subtype="PCM_16")
samples.append({
"id": name[:-4], "filename": name, "audio_path": str((ace_dir / name).resolve()),
"caption": item["caption"], "lyrics": "", "raw_lyrics": "",
"duration": round((end - start) / rate, 3), "language": item["language"],
"is_instrumental": False, "custom_tag": item["trigger"], "labeled": False,
"reviewed": False, "bpm": None, "keyscale": "", "timesignature": "",
})
provenance.append({
"filename": name, "source": entry["path"],
"start_seconds": round(start / rate, 6), "end_seconds": round(end / rate, 6),
})
log(f" {len(ranges)} clips retained")
if not samples:
raise ValueError("No usable clips. Check the audio and silence threshold; source files were not changed.")
path = output / "dataset.json"
write_json(path, {
"metadata": {
"name": item["name"], "custom_tag": item["trigger"], "tag_position": "prepend",
"created_at": now(), "num_samples": len(samples), "all_instrumental": False,
"genre_ratio": 0,
},
"samples": samples, "warnings": warnings,
})
write_json(output / "segments.json", {
"clip_seconds": clip_seconds, "silence_db": silence_db, "sources": provenance,
})
item["dataset"] = str(path.relative_to(WORKSPACE / item["id"]))
save_character(item)
log(f"Prepared {len(samples)} clips. Lyrics and listening review are still required for ACE-Step.")
for warning in warnings:
log("WARNING: " + warning)
return path
def require_review(path: Path) -> dict:
data = read_json(path)
samples = data.get("samples", [])
if not samples:
raise ValueError("The dataset is empty.")
incomplete = [
s["filename"] for s in samples
if not s.get("reviewed") or not s.get("caption", "").strip() or not s.get("lyrics", "").strip()
or s.get("is_instrumental", True)
]
if incomplete:
raise ValueError(
f"Review and annotate {len(incomplete)} clips before ACE-Step preprocessing: "
+ ", ".join(incomplete[:8])
+ ". For wordless singing, describe the actual vocalisation; do not label vocals instrumental."
)
return data
def backend_paths(model: str) -> tuple[Path, Path]:
if model not in MODELS:
raise ValueError("Choose an ACE-Step XL model.")
config = settings()
root, python = Path(config["ace_root"]), Path(config["ace_python"])
for path in (python, root / "train.py", root / "checkpoints" / model / "config.json"):
if not path.is_file():
raise FileNotFoundError(f"Required ACE-Step file not found: {path}")
return root, python
def training_command(item: dict, options: dict, tensor_dir: Path, output: Path) -> list[str]:
model = options["model"]
root, python = backend_paths(model)
turbo = MODELS[model] == "xl_turbo"
return [
str(python), "-u", str(root / "train.py"), "--plain", "--yes", "fixed",
"--checkpoint-dir", str(root / "checkpoints"),
# Literal folder names also work with older upstream XL path validators.
"--model-variant", model, "--base-model", MODELS[model],
"--dataset-dir", str(tensor_dir), "--output-dir", str(output),
"--device", "cuda", "--precision", "bf16", "--num-workers", "0",
"--batch-size", "1", "--gradient-accumulation", str(options["accumulation"]),
"--epochs", str(options["epochs"]), "--lr", str(options["learning_rate"]),
"--rank", str(options["rank"]), "--alpha", str(options["rank"] * 2),
"--save-every", str(min(10, options["epochs"])), "--warmup-steps", "10",
"--gradient-checkpointing", "--offload-encoder", "--optimizer-type", "adamw",
"--shift", "3.0" if turbo else "1.0", "--num-inference-steps", "8" if turbo else "50",
]
def export_svc(item: dict, log: Callable[[str], None]) -> Path:
dataset = dataset_path(item)
data = read_json(dataset)
files = sorted((dataset.parent / "svc_audio").glob("*.wav"))
if not files or len(files) != len(data["samples"]):
raise ValueError("SVC audio is incomplete. Prepare a new dataset.")
target = dataset.parent / f"{slug(item['name'])}-svc-dataset.zip"
with zipfile.ZipFile(target, "w", compression=zipfile.ZIP_STORED) as archive:
for path in files:
archive.write(path, arcname=f"{slug(item['name'])}/{path.name}")
archive.writestr("TRAINING.txt", (
f"Character: {item['name']}\n"
"Dataset: mono 40 kHz PCM-16 WAV; not a trained conversion model.\n\n"
"Recommended separate SVC backend: Applio (RVC).\n"
"https://github.com/IAHispano/Applio\n\n"
"Extract the character folder. In Applio's Train tab, choose a unique model name,\n"
"point the dataset field to the extracted folder, and choose 40 kHz.\n"
"Preprocess, extract pitch/features (RMVPE is a common singing choice),\n"
"then train and build the retrieval index. Review its current training guide\n"
"and install its required base assets in its own environment.\n"
"Use the resulting .pth and .index together for conversion of isolated vocals.\n"
"Preserve pitch unless a deliberate transpose is wanted; mix vocals back into\n"
"the original instrumental. Do not feed a full mix into a vocal-only converter.\n\n"
"An ACE-Step LoRA is not an RVC model. Do not rename adapter files to .pth.\n"
"Train only voices you are authorized to use. Listen to every clip first.\n"
))
log(f"SVC dataset exported: {target}. Train .pth / .index separately in Applio.")
return target
class JobManager:
def __init__(self) -> None:
self.lock = threading.RLock()
self.jobs: list[dict] = []
self.busy = False
self.process: subprocess.Popen | None = None
self.closing = False
def load_history(self) -> None:
with self.lock:
paths = sorted(WORKSPACE.glob("*/jobs/*/job.json"), reverse=True)
loaded = [read_json(path) for path in paths]
self.jobs = sorted(loaded, key=lambda job: job["started_at"])[-30:]
for job in self.jobs:
if job["status"] in {"queued", "running"}:
job.update(status="failed", error="Studio stopped before this job completed.",
ended_at=now())
self.persist(job)
def ensure_idle(self) -> None:
if self.busy:
raise ValueError("A job is running. Wait for it to finish before changing the workspace.")
def start(self, cid: str, action: str, options: dict) -> dict:
with self.lock:
self.ensure_idle()
if self.closing:
raise ValueError("The studio is shutting down.")
item = character(cid)
if action in {"preprocess", "train"} and not options["authorized"]:
raise ValueError("Confirm that you are authorized to train this voice.")
job_id = datetime.now().strftime("%Y%m%d-%H%M%S") + "-" + uuid.uuid4().hex[:8]
job = {
"id": job_id, "character_id": cid, "character_name": item["name"],
"action": action, "status": "queued", "started_at": now(), "ended_at": None,
"error": None, "log": [], "artifacts": [],
"options": {key: value for key, value in options.items() if key != "authorized"},
"authorization_confirmed": bool(options["authorized"]),
}
self.jobs.append(job)
self.busy = True
self.persist(job)
thread = threading.Thread(target=self.run, args=(job, item, options), daemon=True)
thread.start()
return dict(job)
def persist(self, job: dict) -> None:
write_json(WORKSPACE / job["character_id"] / "jobs" / job["id"] / "job.json", job)
def log(self, job: dict, message: str) -> None:
with self.lock:
line = re.sub(r"\x1b\[[0-9;]*[A-Za-z]", "", message.rstrip())
job["log"].append(line)
job["log"] = job["log"][-400:]
path = WORKSPACE / job["character_id"] / "jobs" / job["id"] / "output.log"
with path.open("a", encoding="utf-8") as handle:
handle.write(line + "\n")
def run_process(self, job: dict, command: list[str], cwd: Path) -> None:
self.log(job, subprocess.list2cmdline(command))
environment = {**os.environ, "PYTHONUTF8": "1", "PYTHONUNBUFFERED": "1"}
with self.lock:
if self.closing:
raise RuntimeError("Studio shutdown interrupted the job.")
process = subprocess.Popen(
command, cwd=str(cwd), stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
text=True, encoding="utf-8", errors="replace", env=environment,
)
self.process = process
try:
assert process.stdout is not None
reported_failure = False
for line in process.stdout:
self.log(job, line)
if "[FAIL]" in line:
reported_failure = True
code = process.wait()
if code:
raise RuntimeError(f"Backend exited with code {code}. See the full job log.")
if reported_failure:
raise RuntimeError("Backend reported a failure despite exiting with code 0. See the full job log.")
finally:
with self.lock:
self.process = None
def run(self, job: dict, item: dict, options: dict) -> None:
log = lambda message: self.log(job, message)
try:
with self.lock:
job["status"] = "running"
self.persist(job)
log(f"{job['action'].upper()} | {item['name']} | {now()}")
action = job["action"]
if action == "prepare":
prepare_dataset(item, options["clip_seconds"], options["silence_db"], log)
elif action == "export_svc":
job["artifacts"].append(str(export_svc(item, log).relative_to(WORKSPACE)))
else:
path = dataset_path(item)
data = require_review(path)
model = options["model"]
root, python = backend_paths(model)
if action == "preprocess":
tensor_dir = path.parent / "tensors" / job["id"]
command = [
str(python), "-u", str(Path(__file__).with_name("ace_worker.py")),
"--ace-root", str(root), "--dataset", str(path), "--output", str(tensor_dir),
"--model", model,
]
self.run_process(job, command, root)
count = sum(not p.name.endswith(".tmp.pt") for p in tensor_dir.glob("*.pt"))
if count != len(data["samples"]):
raise RuntimeError(f"Expected {len(data['samples'])} tensors, found {count}.")
write_json(path.parent / "ready.json", {
"dataset_sha256": fingerprint(path), "model": model,
"tensor_dir": str(tensor_dir), "count": count, "created_at": now(),
})
log(f"All {count} clips preprocessed for {model}.")
elif action == "train":
ready_path = path.parent / "ready.json"
if not ready_path.is_file():
raise ValueError("Preprocess this dataset before training.")
ready = read_json(ready_path)
if ready["dataset_sha256"] != fingerprint(path) or ready["model"] != model:
raise ValueError("Dataset or model changed. Preprocess again before training.")
tensor_dir = Path(ready["tensor_dir"])
if sum(not p.name.endswith(".tmp.pt") for p in tensor_dir.glob("*.pt")) != len(data["samples"]):
raise ValueError("Preprocessed tensors are incomplete. Preprocess again.")
output = WORKSPACE / item["id"] / "jobs" / job["id"] / "adapter"
command = training_command(item, options, tensor_dir, output)
write_json(output.parent / "recipe.json", {
"character": item["name"], "trigger": data["metadata"]["custom_tag"], "base_model": model,
"dataset_sha256": fingerprint(path), "command": command,
"authorization_confirmed_at": job["started_at"],
})
self.run_process(job, command, root)
final = output / "final"
weights = sorted(final.rglob("*.safetensors"))
if not weights or not (final / "adapter_config.json").is_file():
raise RuntimeError("Backend did not produce a complete final adapter. Inspect the job log.")
job["artifacts"] = [str(p.relative_to(WORKSPACE)) for p in weights]
log("Adapter files found. Audition checkpoints in the same XL base model before choosing one.")
else:
raise ValueError("Unknown job action.")
with self.lock:
job["status"] = "completed"
except Exception as exc:
self.log(job, traceback.format_exc())
with self.lock:
job["status"], job["error"] = "failed", str(exc)
finally:
with self.lock:
job["ended_at"] = now()
self.busy = False
self.persist(job)
def snapshot(self) -> list[dict]:
with self.lock:
return [json.loads(json.dumps(job)) for job in reversed(self.jobs[-30:])]
def shutdown(self) -> None:
with self.lock:
self.closing = True
process = self.process
if process and process.poll() is None:
process.terminate()
try:
process.wait(timeout=10)
except subprocess.TimeoutExpired:
process.kill()
process.wait()
jobs = JobManager()