from fastapi import APIRouter, File, UploadFile, HTTPException from fastapi.responses import StreamingResponse from pydantic import BaseModel from ai4b import BhashiniClient from fast_langdetect import detect import io import base64 import os router = APIRouter() ULCA_USER_ID = os.getenv("ULCA_USER_ID") ULCA_API_KEY = os.getenv("ULCA_API_KEY") client = BhashiniClient(user_id=ULCA_USER_ID, api_key=ULCA_API_KEY) class TTSRequest(BaseModel): text: str voice: str = "female" SUPPORTED_LANGUAGES = {'pa', 'mr', 'bn', 'en', 'as', 'or', 'ta', 'te', 'kn', 'gu', 'hi', 'ml'} def detect_language(text): text = text.replace("\n", " ") try: result = detect(text, low_memory=False) detected_lang = result['lang'] if detected_lang in SUPPORTED_LANGUAGES: return detected_lang except: pass if any('\u0980' <= char <= '\u09FF' for char in text): return 'brx' elif any('\uABC0' <= char <= '\uABFF' for char in text): return 'mni' return 'en' @router.post("/tts") async def text_to_speech(request: TTSRequest): try: detected_language = detect_language(request.text) tts_result = client.tts( request.text, source_language=detected_language, gender=request.voice ) audio_base64 = tts_result['pipelineResponse'][0]['audio'][0]['audioContent'] audio_data = base64.b64decode(audio_base64) return StreamingResponse(io.BytesIO(audio_data), media_type="audio/wav") except Exception as e: raise HTTPException(status_code=500, detail=str(e)) @router.post("/asr") async def speech_to_text(file: UploadFile = File(...), source_language: str = "en"): try: audio_content = await file.read() asr_result = client.asr(audio_content, source_language=source_language) # Extract the transcribed text from the complex JSON structure transcribed_text = asr_result['pipelineResponse'][0]['output'][0]['source'] return {"text": transcribed_text} except KeyError: raise HTTPException(status_code=500, detail=f"Unexpected response structure from ASR service out:{asr_result}") except Exception as e: raise HTTPException(status_code=500, detail=str(e))