app.py

from fastapi import FastAPI
from fastapi.responses import StreamingResponse
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech
import torch
import soundfile as sf
import io

app = FastAPI()

# Load processor & model
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts")

# Dummy speaker embedding (flat voice). You can later replace with real embeddings.
speaker_embeddings = torch.zeros((1, 512))

@app.get("/speak")
def speak(text: str):
    inputs = processor(text=text, return_tensors="pt")

    # Generate speech
    speech = model.generate_speech(inputs["input_ids"], speaker_embeddings)

    # Save into memory buffer
    buf = io.BytesIO()
    sf.write(buf, speech.numpy(), 16000, format="WAV")
    buf.seek(0)

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