app.py

from fastapi import FastAPI, File, UploadFile, HTTPException
import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
import requests
import json
import tempfile
import os

app = FastAPI()

# Set up Whisper model
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

model_id = "openai/whisper-large-v3-turbo"

model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id)

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    torch_dtype=torch_dtype,
    device=device,
)

OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY", "")
OPENROUTER_URL = "https://openrouter.ai/api/v1/chat/completions"

@app.post("/transcribe-analyze/")
async def transcribe_analyze(file: UploadFile = File(...)):
    try:
        # Save the uploaded file temporarily
        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as temp_audio:
            temp_audio.write(await file.read())
            temp_audio_path = temp_audio.name

        # Transcribe audio
        transcription_result = pipe(temp_audio_path, return_timestamps=True)
        transcription = transcription_result["text"]

        # Send transcription to AI for classification
        response = requests.post(
            url=OPENROUTER_URL,
            headers={
                "Authorization": f"Bearer {OPENROUTER_API_KEY}",
                "Content-Type": "application/json"
            },
            data=json.dumps({
                "model": "meta-llama/llama-3.1-70b-instruct:free",
                "messages": [
                    {
                        "role": "user",
                        "content": f"You are an AI Assistant that is given the transcript between a call agent and a lead, and you must classify if the lead happily agreed to the booking. The response should have 4 parts: 1. Appointment Booked: Yes/No, 2. Short reason for your answer, 3. Short summary of the call, 4. Lead's overall emotion. \n Here is the transcription: {transcription}",
                    }
                ]
            })
        )

        ai_response = response.json().get("choices", [{}])[0].get("message", {}).get("content", "No response from AI.")

        # Remove temporary file
        os.remove(temp_audio_path)

        return {"transcription": transcription, "ai_response": ai_response}

    except Exception as e:
        return HTTPException(status_code=500, detail=str(e))