Deploy XTTS fallback as HF Inference Endpoint

Dockerfile

@@ -0,0 +1,27 @@
+FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime
+
+# Prevent interactive tzdata prompts and accept Coqui TOS
+ENV DEBIAN_FRONTEND=noninteractive
+ENV TZ=Etc/UTC
+ENV COQUI_TOS_AGREED=1
+
+# Install system dependencies
+RUN apt-get update && \
+    apt-get install -y git ffmpeg libsndfile1 tzdata && \
+    rm -rf /var/lib/apt/lists/*
+
+# Create working directory
+WORKDIR /app
+
+# Install Python dependencies
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy app code
+COPY app.py .
+
+# Expose port
+EXPOSE 7860
+
+# Start FastAPI server
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -0,0 +1,24 @@
+from fastapi import FastAPI, UploadFile, Form
+from fastapi.responses import FileResponse
+import uuid, os, soundfile as sf
+from TTS.api import TTS
+
+app = FastAPI()
+
+# Load XTTS model (multi-lingual, multi-speaker)
+# This will be slow on cold start but stays cached in memory
+tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to("cuda")
+
+@app.post("/generate")
+async def generate(text: str = Form(...), speaker_wav: UploadFile = None):
+    output_path = f"/tmp/{uuid.uuid4().hex}.wav"
+    
+    if speaker_wav:
+        wav_path = f"/tmp/{speaker_wav.filename}"
+        with open(wav_path, "wb") as f:
+            f.write(await speaker_wav.read())
+        tts.tts_to_file(text=text, speaker_wav=wav_path, file_path=output_path)
+    else:
+        tts.tts_to_file(text=text, speaker="random", file_path=output_path)
+
+    return FileResponse(output_path, media_type="audio/wav")

huggingface.yaml

@@ -0,0 +1,15 @@
+# Hugging Face Inference Endpoint config
+build:
+  dockerfile: Dockerfile
+
+resources:
+  # Use CPU instance for free/low-cost deployment
+  cpu: "2"
+  memory: "8Gi"
+
+scaling:
+  # Scale-to-zero when idle → no charges while unused
+  min_replicas: 0
+  max_replicas: 1
+
+entrypoint: uvicorn app:app --host 0.0.0.0 --port 7860

requirements.txt

@@ -0,0 +1,8 @@
+fastapi
+uvicorn
+soundfile
+numpy
+torch
+torchaudio
+transformers
+TTS==0.21.1