main.py

from fastapi import FastAPI, File, UploadFile, Form
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
from pydantic import BaseModel
from transformers import pipeline, MarianMTModel, MarianTokenizer, WhisperProcessor, WhisperForConditionalGeneration
import torch
import tempfile
import soundfile as sf

app = FastAPI()

# Allow frontend to call backend
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Supported languages
translation_models = {
    "fr": "Helsinki-NLP/opus-mt-en-fr",
    "es": "Helsinki-NLP/opus-mt-en-es",
    "de": "Helsinki-NLP/opus-mt-en-de",
    "it": "Helsinki-NLP/opus-mt-en-it",
    "hi": "Helsinki-NLP/opus-mt-en-hi",
    "ru": "Helsinki-NLP/opus-mt-en-ru",
    "zh": "Helsinki-NLP/opus-mt-en-zh",
    "ar": "Helsinki-NLP/opus-mt-en-ar",
    "ta": "Helsinki-NLP/opus-mt-en-ta"
}

# Load models once
generator = pipeline("text-generation", model="distilgpt2", framework="tf", from_tf=True)
whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")
whisper_processor = WhisperProcessor.from_pretrained("openai/whisper-base")

@app.get("/")
def root():
    return {"message": "Backend is live ✅"}

@app.get("/generate")
def generate_and_translate(prompt: str, target_lang: str):
    try:
        if target_lang not in translation_models:
            return {"error": "Unsupported language."}
        
        # 1. Generate English sentence
        result = generator(prompt, max_length=30, num_return_sequences=1)[0]["generated_text"]
        english_sentence = result.strip()

        # 2. Translate
        model_name = translation_models[target_lang]
        tokenizer = MarianTokenizer.from_pretrained(model_name)
        model = MarianMTModel.from_pretrained(model_name)
        tokens = tokenizer(english_sentence, return_tensors="pt", padding=True)
        translated_ids = model.generate(**tokens)
        translated_text = tokenizer.decode(translated_ids[0], skip_special_tokens=True)

        return {"english": english_sentence, "translated": translated_text}
    except Exception as e:
        return JSONResponse(status_code=500, content={"error": str(e)})

class TranslateRequest(BaseModel):
    text: str
    target_lang: str

@app.post("/translate")
def translate_text(data: TranslateRequest):
    try:
        if data.target_lang not in translation_models:
            return {"error": "Unsupported language."}
        
        model_name = translation_models[data.target_lang]
        tokenizer = MarianTokenizer.from_pretrained(model_name)
        model = MarianMTModel.from_pretrained(model_name)
        tokens = tokenizer(data.text, return_tensors="pt", padding=True)
        translated_ids = model.generate(**tokens)
        translated_text = tokenizer.decode(translated_ids[0], skip_special_tokens=True)

        return {"translated_text": translated_text}
    except Exception as e:
        return JSONResponse(status_code=500, content={"error": str(e)})

@app.post("/transcribe")
async def transcribe_audio(audio: UploadFile = File(...)):
    try:
        temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
        temp_file.write(await audio.read())
        temp_file.close()

        audio_data, _ = sf.read(temp_file.name)
        inputs = whisper_processor(audio_data, sampling_rate=16000, return_tensors="pt")
        predicted_ids = whisper_model.generate(inputs["input_features"])
        transcription = whisper_processor.decode(predicted_ids[0], skip_special_tokens=True)

        return {"transcribed_text": transcription}
    except Exception as e:
        return JSONResponse(status_code=500, content={"error": str(e)})

@app.post("/process")
async def transcribe_and_translate_audio(
    audio: UploadFile = File(...),
    target_lang: str = Form(...)
):
    try:
        if target_lang not in translation_models:
            return {"error": "Unsupported language."}

        # Save uploaded file
        temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
        temp_file.write(await audio.read())
        temp_file.close()

        # Transcribe
        audio_data, _ = sf.read(temp_file.name)
        inputs = whisper_processor(audio_data, sampling_rate=16000, return_tensors="pt")
        predicted_ids = whisper_model.generate(inputs["input_features"])
        transcription = whisper_processor.decode(predicted_ids[0], skip_special_tokens=True)

        # Translate
        model_name = translation_models[target_lang]
        tokenizer = MarianTokenizer.from_pretrained(model_name)
        model = MarianMTModel.from_pretrained(model_name)
        tokens = tokenizer(transcription, return_tensors="pt", padding=True)
        translated_ids = model.generate(**tokens)
        translated_text = tokenizer.decode(translated_ids[0], skip_special_tokens=True)

        return {
            "transcribed_text": transcription,
            "translated_text": translated_text
        }
    except Exception as e:
        return JSONResponse(status_code=500, content={"error": str(e)})

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)