app.py

from fastapi import FastAPI, Form
from fastapi.responses import JSONResponse, FileResponse
import uvicorn
from pydantic import BaseModel
import numpy as np
import io
import soundfile as sf
import base64
from asr import transcribe, ASR_LANGUAGES
from tts import synthesize, TTS_LANGUAGES
from lid import identify

app = FastAPI(title="MMS: Scaling Speech Technology to 1000+ languages")

class TTSRequest(BaseModel):
    text: str
    language: str
    speed: float

class AudioRequest(BaseModel):
    audio: str  # Base64 encoded audio data
    language: str

@app.post("/transcribe")
async def transcribe_audio(request: AudioRequest):
    audio_bytes = base64.b64decode(request.audio)
    audio_array, sample_rate = sf.read(io.BytesIO(audio_bytes))
    
    result = transcribe(audio_array, request.language)
    return JSONResponse(content={"transcription": result})

@app.post("/synthesize")
async def synthesize_speech(request: TTSRequest):
    audio, filtered_text = synthesize(request.text, request.language, request.speed)
    
    # Convert numpy array to bytes
    buffer = io.BytesIO()
    sf.write(buffer, audio, 22050, format='wav')
    buffer.seek(0)
    
    return FileResponse(
        buffer, 
        media_type="audio/wav", 
        headers={"Content-Disposition": "attachment; filename=synthesized_audio.wav"}
    )

@app.post("/identify")
async def identify_language(request: AudioRequest):
    audio_bytes = base64.b64decode(request.audio)
    audio_array, sample_rate = sf.read(io.BytesIO(audio_bytes))
    
    result = identify(audio_array)
    return JSONResponse(content={"language_identification": result})

@app.get("/asr_languages")
async def get_asr_languages():
    return JSONResponse(content=ASR_LANGUAGES)

@app.get("/tts_languages")
async def get_tts_languages():
    return JSONResponse(content=TTS_LANGUAGES)