app.py

import gradio as gr
import torch
from transformers import pipeline, WhisperProcessor, WhisperForConditionalGeneration, BartForConditionalGeneration, BartTokenizer
from huggingface_hub import login
import os

# Retrieve the token from the environment variable
hf_api_token = os.getenv("HF_API_TOKEN")

if hf_api_token is None:
    raise ValueError("HF_API_TOKEN environment variable is not set")

# Authenticate with Hugging Face
login(token=hf_api_token, add_to_git_credential=True)

# Initialize the Whisper processor and model
whisper_processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")

# Initialize the summarization model and tokenizer
# Use BART model for summarization
summarization_model = BartForConditionalGeneration.from_pretrained("facebook/bart-large-cnn")
summarization_tokenizer = BartTokenizer.from_pretrained("facebook/bart-large-cnn")

# Function to transcribe audio
def transcribe_audio(audio_file):
    # Load audio file
    audio_input, _ = whisper_processor(audio_file, return_tensors="pt", sampling_rate=16000).input_values
    # Generate transcription
    transcription_ids = whisper_model.generate(audio_input)
    transcription = whisper_processor.decode(transcription_ids[0])
    return transcription

# Function to summarize text
def summarize_text(text):
    inputs = summarization_tokenizer(text, return_tensors="pt", max_length=1024, truncation=True)
    summary_ids = summarization_model.generate(inputs.input_ids, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
    summary