app.py

from fastapi import FastAPI, Response
from fastapi.responses import FileResponse
from kokoro import KPipeline
import soundfile as sf
import os

app = FastAPI()

# Initialize pipeline once at startup
pipeline = KPipeline(lang_code='a')

@app.post("/generate")
async def generate_audio(text: str, voice: str = "af_heart", speed: float = 1.0):
    # Generate audio
    generator = pipeline(
        text, 
        voice=voice,
        speed=speed,
        split_pattern=r'\n+'
    )
    
    # # Save first segment only for demo
    # for i, (gs, ps, audio) in enumerate(generator):
    #     sf.write(f"output_{i}.wav", audio, 24000)
    #     return FileResponse(
    #         f"output_{i}.wav",
    #         media_type="audio/wav",
    #         filename="output.wav"
    #     )
    
    # return Response("No audio generated", status_code=400)


    # Process only the first segment for demo
    for i, (gs, ps, audio) in enumerate(generator):
        # Convert to 16-bit PCM
        # Ensure the audio is in the range [-1, 1] first
        audio = np.clip(audio, -1, 1)
        # Convert to 16-bit signed integers
        pcm_data = (audio * 32767).astype(np.int16)
        
        # Convert to bytes, ensuring little-endian
        raw_audio = pcm_data.tobytes(order='C')
        
        # Return PCM data with appropriate headers
        return Response(
            content=raw_audio,
            media_type="audio/l16",  # Linear PCM
            headers={
                "Content-Disposition": f'attachment; filename="output.pcm"',
                "X-Sample-Rate": "24000",
                "X-Bits-Per-Sample": "16",
                "X-Channels": "1",
                "X-Encoding": "signed-integer",
                "X-Endian": "little"
            }
        )
    return Response("No audio generated", status_code=400)