Upload app.py with huggingface_hub

app.py

@@ -0,0 +1,58 @@
+import io, base64, tempfile, os
+from fastapi import FastAPI
+from fastapi.responses import StreamingResponse
+from pydantic import BaseModel
+import torchaudio as ta
+from chatterbox.tts import ChatterboxTTS
+
+app = FastAPI()
+
+# Load model once at startup
+print("Loading ChatterboxTTS model...")
+model = ChatterboxTTS.from_pretrained(device="cpu")
+print("Model ready.")
+
+
+class TTSRequest(BaseModel):
+    text: str
+    ref_audio: str | None = None        # base64-encoded audio file
+    exaggeration: float = 0.5
+    cfg_weight: float = 0.5
+    temperature: float = 0.8
+
+
+@app.get("/health")
+def health():
+    return {"status": "ok"}
+
+
+@app.post("/synthesize")
+def synthesize(req: TTSRequest):
+    ref_path = None
+
+    # Write ref audio to a temp file if provided
+    if req.ref_audio:
+        audio_bytes = base64.b64decode(req.ref_audio)
+        tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
+        tmp.write(audio_bytes)
+        tmp.close()
+        ref_path = tmp.name
+
+    try:
+        wav = model.generate(
+            req.text,
+            audio_prompt_path=ref_path,
+            exaggeration=req.exaggeration,
+            cfg_weight=req.cfg_weight,
+            temperature=req.temperature,
+        )
+    finally:
+        if ref_path and os.path.exists(ref_path):
+            os.unlink(ref_path)
+
+    # Write wav to buffer and return as audio/wav
+    buf = io.BytesIO()
+    ta.save(buf, wav, model.sr, format="wav")
+    buf.seek(0)
+
+    return StreamingResponse(buf, media_type="audio/wav")