feat: Supertonic 엔진 추가(자연스러운 한국어, MIT, ONNX)

- 전용 venv 워커(8002)로 격리, MeloTTS(8788)와 병행
- 남녀 10종 음성, 한국어/자동감지(다국어), 속도·피치
- 모델 ~260MB 베이킹, torch/GPU 불필요
- entrypoint로 워커+메인앱 동시 기동

supertonic_worker/requirements.txt

@@ -0,0 +1,4 @@
+# Supertonic (수퍼톤) — 자연스러운 다국어 TTS, ONNX 런타임, torch 불필요
+supertonic==1.3.1
+fastapi==0.115.6
+uvicorn==0.30.0

supertonic_worker/worker.py

@@ -0,0 +1,91 @@
+"""Supertonic 합성 워커 (전용 venv에서 실행).
+
+- ONNX 런타임 기반, torch 불필요, 모델 ~260MB
+- 다국어(31개) 지원. 한국어는 lang="ko", 자동감지는 lang="na"
+- 프리셋 음성 10종(M1~M5, F1~F5)
+- 속도(speed) 네이티브, 피치(pitch)는 librosa 후처리
+"""
+from __future__ import annotations
+
+import io
+
+import numpy as np
+import soundfile as sf
+
+try:
+    import librosa
+except Exception:  # pragma: no cover
+    librosa = None
+
+from fastapi import FastAPI
+from fastapi.responses import Response
+from pydantic import BaseModel
+
+from supertonic import TTS
+
+VOICES = ["M1", "M2", "M3", "M4", "M5", "F1", "F2", "F3", "F4", "F5"]
+SR = 44100
+
+_tts = None
+_styles: dict = {}
+
+
+def get_tts() -> TTS:
+    global _tts
+    if _tts is None:
+        _tts = TTS(auto_download=True)
+    return _tts
+
+
+def get_style(name: str):
+    if name not in _styles:
+        _styles[name] = get_tts().get_voice_style(voice_name=name)
+    return _styles[name]
+
+
+app = FastAPI(title="Supertonic Worker")
+
+
+class SynthReq(BaseModel):
+    text: str
+    voice: str = "F1"
+    lang: str = "ko"
+    speed: float = 1.0
+    pitch: float = 0.0
+
+
+@app.on_event("startup")
+def _startup() -> None:
+    get_tts()
+    get_style("F1")
+
+
+@app.get("/health")
+def health() -> dict:
+    get_tts()
+    return {"status": "ok", "voices": VOICES, "sr": SR}
+
+
+@app.post("/synth")
+def synth(req: SynthReq) -> Response:
+    text = (req.text or "").strip()
+    if not text:
+        return Response(content="empty text", status_code=400)
+
+    tts = get_tts()
+    voice = req.voice if req.voice in VOICES else "F1"
+    style = get_style(voice)
+    lang = req.lang if req.lang in ("ko", "na", "en", "ja", "zh") else "ko"
+    speed = float(max(0.5, min(2.0, req.speed)))
+    pitch = float(max(-12.0, min(12.0, req.pitch)))
+
+    wav, _dur = tts.synthesize(text, voice_style=style, lang=lang, speed=speed)
+    audio = np.asarray(wav).reshape(-1).astype(np.float32)
+
+    if abs(pitch) > 1e-3 and librosa is not None:
+        audio = librosa.effects.pitch_shift(audio, sr=SR, n_steps=pitch)
+
+    buf = io.BytesIO()
+    sf.write(buf, audio, SR, format="WAV", subtype="PCM_16")
+    buf.seek(0)
+    return Response(content=buf.read(), media_type="audio/wav")