app.py

import gradio as gr
import whisper
from transformers import pipeline

# Load Whisper model
whisper_model = whisper.load_model("base")

# Load traditional summarization models
def get_summarizer(model_name):
    if model_name == "BART (facebook/bart-large-cnn)":
        return pipeline("summarization", model="facebook/bart-large-cnn")
    elif model_name == "T5 (t5-small)":
        return pipeline("summarization", model="t5-small")
    elif model_name == "Pegasus (google/pegasus-xsum)":
        return pipeline("summarization", model="google/pegasus-xsum")
    else:
        return None

# Function to transcribe audio file using Whisper
def transcribe_audio(model_size, audio):
    model = whisper.load_model(model_size)
    result = model.transcribe(audio)
    transcription = result['text']
    return transcription

# Function to summarize the transcribed text
def summarize_text(transcription, model_name):
    if len(transcription.strip()) == 0:
        return "No text to summarize."
    
    summarizer = get_summarizer(model_name)
    
    if summarizer:
        summary = summarizer(transcription, max_length=150, min_length=30, do_sample=False)[0]['summary_text']
        return summary
    else:
        return "Invalid summarization model selected."

# Create a Gradio interface that combines transcription and summarization
def combined_transcription_and_summarization(model_size, summarizer_model, audio):
    # Step 1: Transcribe the audio using Whisper
    transcription = transcribe_audio(model_size, audio)
    
    # Step 2: Summarize the transcribed text using the chosen summarizer model
    summary = summarize_text(transcription, summarizer_model)
    
    return transcription, summary

# Gradio interface for transcription and summarization
iface = gr.Interface(
    fn=combined_transcription_and_summarization,   # The combined function
    inputs=[
        gr.Dropdown(label="Choose Whisper Model", choices=["tiny", "base", "small", "medium", "large"], value="base"),  # Whisper model selection
        gr.Dropdown(label="Choose Summarizer Model", choices=["BART (facebook/bart-large-cnn)", "T5 (t5-small)", "Pegasus (google/pegasus-xsum)"], value="BART (facebook/bart-large-cnn)"),  # Summarizer model selection
        gr.Audio(type="filepath")  # Audio upload
    ],
    outputs=[
        gr.Textbox(label="Transcription"),         # Output for the transcribed text
        gr.Textbox(label="Summary")                # Output for the summary
    ],
    title="Whisper Audio Transcription and Summarization",
    description="Upload an audio file, choose a Whisper model for transcription, and a summarization model to summarize the transcription."
)

# Launch the interface
iface.launch()