app.py

import os

# Redirect Hugging Face cache to writable /tmp
os.environ["HF_HOME"] = "/tmp/hf_cache"
os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf_cache"

from fastapi import FastAPI, Query
from fastapi.responses import StreamingResponse
import io
import torch
import soundfile as sf
from kokoro import KPipeline

app = FastAPI(title="Kokoro TTS API")

# Initialize pipeline once
pipeline = KPipeline(lang_code='a')  # default English

@app.post("/tts")
async def tts_endpoint(text: str = Query(..., min_length=1)):
    """
    Convert text to speech and stream WAV.
    """
    try:
        # Generate audio chunks
        generator = pipeline(text, voice='af_heart', speed=1.0, split_pattern=r'\n+')
        audio_list = [audio for _, _, _, audio in generator]
        audio_tensor = torch.cat(audio_list, dim=0)
    except Exception as e:
        return {"error": f"TTS generation failed: {e}"}

    # Convert to WAV in-memory
    buffer = io.BytesIO()
    sf.write(buffer, audio_tensor.numpy(), 24000, format="WAV")
    buffer.seek(0)

    return StreamingResponse(buffer, media_type="audio/wav")