app.py

from fastapi import FastAPI
from fastapi.responses import StreamingResponse
from transformers import pipeline
import io
import soundfile as sf

app = FastAPI()

# Load TTS pipeline
tts = pipeline("text-to-speech", model="suno/bark-small")

@app.get("/speak")
def speak(text: str):
    # Generate speech (numpy float32 array)
    output = tts(text)
    audio = output["audio"]

    # Write WAV to in-memory buffer
    buf = io.BytesIO()
    sf.write(buf, audio, 24000, format="WAV")  # Bark uses 24kHz
    buf.seek(0)

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