app.py

import gradio as gr
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC, AutoTokenizer, BartForConditionalGeneration
import torch
import librosa

# Load BART tokenizer and model for summarization
tokenizer = AutoTokenizer.from_pretrained("facebook/bart-large-cnn")
summarizer = BartForConditionalGeneration.from_pretrained("facebook/bart-large-cnn")

# Load Wav2Vec2 processor and model for transcription
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")

# Check if CUDA is available
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
summarizer.to(device)

def transcribe_and_summarize(audioFile):
    # Load audio as an array
    audio, sampling_rate = librosa.load(audioFile, sr=16000)  # Ensure it's 16kHz for Wav2Vec2
    values = processor(audio, sampling_rate=sampling_rate, return_tensors="pt").input_values

    # Move tensors to GPU if available
    values = values.to(device)

    # Transcription
    with torch.no_grad():  
        logits = model(values).logits
        predictedIDs = torch.argmax(logits, dim=-1)
        transcription = processor.batch_decode(predictedIDs, skip_special_tokens=True)[0]

    # Summarization
    inputs = tokenizer(transcription, return_tensors="pt", truncation=True, max_length=1024)
    inputs = inputs.to(device)  # Move inputs to GPU

    result = summarizer.generate(
        inputs["input_ids"],
        min_length=10,
        max_length=1024,
        no_repeat_ngram_size=2,
        encoder_no_repeat_ngram_size=2,
        repetition_penalty=2.0,
        num_beams=4,
        early_stopping=True,
    )
    summary = tokenizer.decode(result[0], skip_special_tokens=True)

    return transcription, summary

# Gradio interface
iface = gr.Interface(
    fn=transcribe_and_summarize,
    inputs=gr.Audio(type="filepath", label="Upload Audio"),
    outputs=[gr.Textbox(label="Transcription"), gr.Textbox(label="Summary")], 
    title="Audio Transcription and Summarization",
    description="Transcribe and summarize audio using Wav2Vec2 and BART.",
)

iface.launch()