Add XTTS-v2 as first TTS backend, extensible for more

Mirrors the ASR backend pattern exactly: a BACKENDS dict in
tts/router.py maps a backend name to its worker URL, so adding the
next backend is just a new worker file + venv + supervisord entry +
one dict line. XTTS-v2 runs in its own venv (--system-site-packages,
inherits base image torch/CUDA) as a new supervisord program on
port 8005.

XTTS is zero-shot voice cloning, so a reference-voice library was
added (/tts/voices CRUD, stored under /srv/tts/voices) — synthesis
requires picking a previously uploaded voice. Results and model
cache live under /srv/tts/{results,models-cache}, new quadlet
volumes, owned by the same 983:983 user as the existing /srv/asr
dirs.

Fixed two environment issues uncovered while getting XTTS to
actually run inside the container (non-root user, root-built venvs):
- coqui-tts only pins transformers>=4.57 with no ceiling, so pip
  installed an incompatible 5.x; pinned to the last 4.x release.
- HOME defaults to /app (owned by root) for the container's runtime
  user, so numba/matplotlib/torch cache writes failed; HOME is now
  forced to /tmp in all three workers (faster_whisper, qwen3, xtts)
  before any of those libraries get imported.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/tts/workers/xtts_worker.py

@@ -0,0 +1,117 @@
+from __future__ import annotations
+
+import argparse
+import os
+import tempfile
+from pathlib import Path
+from typing import Any, Dict
+
+import sys
+sys.path.insert(0, "/app")
+from tts.config import TTS_MODEL_CACHE, ensure_runtime_dirs
+
+# TTS.api를 import하기 전에 설정해야 적용된다.
+os.environ.setdefault("TTS_HOME", str(TTS_MODEL_CACHE))
+os.environ.setdefault("COQUI_TOS_AGREED", "1")  # XTTS(CPML 라이선스) 동의 프롬프트를 비대화식으로 통과
+# 컨테이너는 983 유저로 실행되는데 HOME(/app)이 root 소유라 numba/matplotlib/torch가
+# 각자 ~/.cache, ~/.config 아래에 쓰려다 실패한다. HOME을 쓰기 가능한 곳으로 돌린다.
+os.environ["HOME"] = "/tmp"  # 컨테이너 기본 HOME=/app은 983 유저가 쓰기 불가 (setdefault로는 덮어쓰기 안 됨)
+os.environ.setdefault("NUMBA_CACHE_DIR", "/tmp/numba_cache")
+
+import soundfile as sf
+import uvicorn
+from fastapi import FastAPI, Form, HTTPException
+from fastapi.responses import Response
+
+app = FastAPI(title="TTS XTTS Worker")
+
+MODEL_NAME = "tts_models/multilingual/multi-dataset/xtts_v2"
+_MODEL_CACHE: Dict[str, Any] = {}
+
+
+def _log(msg: str) -> None:
+    print(f"[xtts] {msg}", flush=True)
+
+
+def _device() -> str:
+    try:
+        import torch
+        return "cuda" if torch.cuda.is_available() else "cpu"
+    except Exception:
+        return "cpu"
+
+
+def _load_model() -> Any:
+    if MODEL_NAME in _MODEL_CACHE:
+        return _MODEL_CACHE[MODEL_NAME]
+
+    from TTS.api import TTS
+
+    _log(f"loading model {MODEL_NAME}")
+    tts = TTS(MODEL_NAME).to(_device())
+    _MODEL_CACHE[MODEL_NAME] = tts
+    _log("model loaded")
+    return tts
+
+
+@app.on_event("startup")
+def startup() -> None:
+    ensure_runtime_dirs()
+
+
+@app.get("/health")
+def health() -> Dict[str, Any]:
+    return {"status": "ok", "device": _device(), "loaded": MODEL_NAME in _MODEL_CACHE}
+
+
+@app.post("/synthesize")
+def synthesize(
+    text: str = Form(...),
+    language: str = Form("ko"),
+    speaker_wav: str = Form(...),
+) -> Response:
+    speaker_path = Path(speaker_wav)
+    if not speaker_path.exists():
+        raise HTTPException(status_code=400, detail=f"speaker_wav not found: {speaker_wav}")
+
+    try:
+        tts = _load_model()
+    except Exception as e:
+        _log(f"model load error: {type(e).__name__}: {e}")
+        raise HTTPException(status_code=500, detail=f"모델 로드 실패: {type(e).__name__}: {e}")
+
+    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
+        out_path = Path(tmp.name)
+
+    try:
+        _log(f"synthesizing language={language} chars={len(text)}")
+        tts.tts_to_file(
+            text=text,
+            speaker_wav=str(speaker_path),
+            language=language,
+            file_path=str(out_path),
+        )
+        audio_bytes = out_path.read_bytes()
+        duration = sf.info(str(out_path)).duration
+        return Response(
+            content=audio_bytes,
+            media_type="audio/wav",
+            headers={"x-audio-duration": str(round(duration, 3))},
+        )
+    except HTTPException:
+        raise
+    except Exception as e:
+        _log(f"synthesize error: {type(e).__name__}: {e}")
+        raise HTTPException(status_code=500, detail=f"{type(e).__name__}: {e}")
+    finally:
+        out_path.unlink(missing_ok=True)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--host", default="0.0.0.0")
+    parser.add_argument("--port", type=int, default=8005)
+    args = parser.parse_args()
+    ensure_runtime_dirs()
+    _log(f"starting host={args.host} port={args.port} device={_device()}")
+    uvicorn.run(app, host=args.host, port=args.port)