diff --git a/Dockerfile b/Dockerfile
index ea94c59..eb35213 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -61,9 +61,22 @@ COPY --from=builder /usr/local/lib/python3.11/site-packages /usr/local/lib/pytho
 COPY --from=builder /usr/local/bin /usr/local/bin
 COPY --from=builder /models /models
 
+# ===== Supertonic 엔진 (자연스러운 한국어, MIT, ONNX, torch 불필요) =====
+# 전용 venv 로 의존성 격리(메인 melo 스택의 librosa 0.9.1 과 충돌 방지).
+COPY supertonic_worker /opt/supertonic_worker
+RUN python -m venv /opt/supertonic-venv \
+ && /opt/supertonic-venv/bin/pip install --no-cache-dir -r /opt/supertonic_worker/requirements.txt
+
+# 모델 사전 다운로드(베이킹, ~260MB) — HF_HOME=/models 캐시에 저장
+RUN HF_HOME=/models /opt/supertonic-venv/bin/python -c "from supertonic import TTS; TTS(auto_download=True)"
+
+ENV SUPERTONIC_URL=http://127.0.0.1:8002
+
 COPY app ./app
 COPY frontend ./frontend
+COPY entrypoint.sh /entrypoint.sh
+RUN chmod +x /entrypoint.sh
 
-# 기본 MeloTTS 단일 엔진 (자동 언어 감지: 한/영 혼합)
+# MeloTTS(8788, 메인) + Supertonic 워커(8002) 동시 기동
 EXPOSE 8788
-CMD ["uvicorn", "app.server:app", "--host", "0.0.0.0", "--port", "8788"]
+CMD ["/entrypoint.sh"]
diff --git a/app/engines/__init__.py b/app/engines/__init__.py
index 3f1a158..b4fdb0a 100644
--- a/app/engines/__init__.py
+++ b/app/engines/__init__.py
@@ -31,7 +31,15 @@ def _build() -> None:
         except Exception as exc:  # pragma: no cover
             print(f"[registry] MeloTTS 로드 실패: {exc}")
 
-        # (CosyVoice 엔진은 사용자 요청에 따라 비활성화 — 기본 MeloTTS만 유지)
+        # Supertonic (자연스러운 한국어, 전용 워커 URL 이 설정된 경우에만)
+        try:
+            from .supertonic_engine import SupertonicEngine
+
+            if SupertonicEngine.available():
+                e = SupertonicEngine()
+                _ENGINES[e.id] = e
+        except Exception as exc:  # pragma: no cover
+            print(f"[registry] Supertonic 엔진 스킵: {exc}")
 
         _INITED = True
 
diff --git a/app/engines/supertonic_engine.py b/app/engines/supertonic_engine.py
new file mode 100644
index 0000000..61f2739
--- /dev/null
+++ b/app/engines/supertonic_engine.py
@@ -0,0 +1,94 @@
+"""Supertonic 엔진 (프록시).
+
+실제 합성은 전용 venv 워커(worker.py)가 담당하고, 이 엔진은 HTTP로 요청을
+전달한다. 워커 URL(SUPERTONIC_URL)이 설정된 경우에만 활성화된다.
+자연스러운 한국어 음성(수퍼톤), ONNX 기반이라 GPU 불필요.
+"""
+from __future__ import annotations
+
+import json
+import os
+import urllib.error
+import urllib.request
+
+from .base import BaseEngine
+
+WORKER_URL = os.environ.get("SUPERTONIC_URL", "").rstrip("/")
+
+_VOICES = [
+    {"id": "F1", "label": "여성 1"},
+    {"id": "F2", "label": "여성 2"},
+    {"id": "F3", "label": "여성 3"},
+    {"id": "F4", "label": "여성 4"},
+    {"id": "F5", "label": "여성 5"},
+    {"id": "M1", "label": "남성 1"},
+    {"id": "M2", "label": "남성 2"},
+    {"id": "M3", "label": "남성 3"},
+    {"id": "M4", "label": "남성 4"},
+    {"id": "M5", "label": "남성 5"},
+]
+
+
+class SupertonicEngine(BaseEngine):
+    id = "supertonic"
+    label = "Supertonic (자연 음성)"
+    license = "MIT"
+    uses_gpu = False
+    notes = "수퍼톤의 자연스러운 다국어 음성. 한국어/자동감지, 남녀 10종, 속도·피치."
+
+    @staticmethod
+    def available() -> bool:
+        return bool(WORKER_URL)
+
+    def describe(self) -> dict:
+        return {
+            "id": self.id,
+            "label": self.label,
+            "license": self.license,
+            "uses_gpu": self.uses_gpu,
+            "notes": self.notes,
+            "languages": [
+                {"code": "KO", "label": "한국어", "speakers": _VOICES},
+                {"code": "AUTO", "label": "자동 감지 (다국어)", "speakers": _VOICES},
+            ],
+            "default_language": "KO",
+            "supports": {"speed": True, "pitch": True},
+        }
+
+    def synth_wav(
+        self,
+        text: str,
+        language: str,
+        speaker: str | None = None,
+        speed: float = 1.0,
+        pitch: float = 0.0,
+    ) -> bytes:
+        text = (text or "").strip()
+        if not text:
+            raise ValueError("텍스트가 비어 있습니다.")
+        if not WORKER_URL:
+            raise ValueError("Supertonic 워커가 설정되지 않았습니다.")
+
+        lang = "na" if (language or "KO").upper() == "AUTO" else "ko"
+        payload = json.dumps(
+            {
+                "text": text,
+                "voice": speaker or "F1",
+                "lang": lang,
+                "speed": speed,
+                "pitch": pitch,
+            }
+        ).encode("utf-8")
+        request = urllib.request.Request(
+            WORKER_URL + "/synth",
+            data=payload,
+            headers={"Content-Type": "application/json"},
+        )
+        try:
+            with urllib.request.urlopen(request, timeout=120) as resp:
+                return resp.read()
+        except urllib.error.HTTPError as exc:
+            detail = exc.read().decode("utf-8", errors="ignore")
+            raise ValueError(f"Supertonic 합성 실패: {detail}")
+        except Exception as exc:  # pragma: no cover
+            raise ValueError(f"Supertonic 워커 연결 실패: {exc}")
diff --git a/entrypoint.sh b/entrypoint.sh
index 50dcd76..0ef5a0f 100644
--- a/entrypoint.sh
+++ b/entrypoint.sh
@@ -1,13 +1,13 @@
 #!/usr/bin/env bash
 set -e
 
-# CosyVoice 워커(전용 venv)를 백그라운드로 기동 — 모델을 1회 로드해 상주
-if [ -n "$COSYVOICE_URL" ] && [ -x /opt/cosyvoice-venv/bin/python ]; then
-  echo "[entrypoint] starting CosyVoice worker on :8001"
-  /opt/cosyvoice-venv/bin/python -m uvicorn worker:app \
-      --app-dir /opt/cosyvoice_worker \
-      --host 127.0.0.1 --port 8001 \
-      > /tmp/cosyvoice_worker.log 2>&1 &
+# Supertonic 워커(전용 venv)를 백그라운드로 기동 — 모델을 1회 로드해 상주
+if [ -n "$SUPERTONIC_URL" ] && [ -x /opt/supertonic-venv/bin/python ]; then
+  echo "[entrypoint] starting Supertonic worker on :8002"
+  /opt/supertonic-venv/bin/python -m uvicorn worker:app \
+      --app-dir /opt/supertonic_worker \
+      --host 127.0.0.1 --port 8002 \
+      > /tmp/supertonic_worker.log 2>&1 &
 fi
 
 # 메인 앱(글로벌 파이썬, MeloTTS 인프로세스)
diff --git a/supertonic_worker/requirements.txt b/supertonic_worker/requirements.txt
new file mode 100644
index 0000000..c083203
--- /dev/null
+++ b/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
diff --git a/supertonic_worker/worker.py b/supertonic_worker/worker.py
new file mode 100644
index 0000000..1007670
--- /dev/null
+++ b/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")