app.py

# app.py
import gradio as gr
from transformers import pipeline
import torch
import tempfile
import os
from TTS.api import TTS
import whisper

# Load question-generation pipeline (use a lightweight model)
qg_pipeline = pipeline("text2text-generation", model="valhalla/t5-small-e2e-qg")

# Load TTS model
tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False)

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

# Generate question and audio from input text
def generate_question(text):
    output = qg_pipeline("generate question: " + text, max_length=64, clean_up_tokenization_spaces=True)[0]['generated_text']
    
    # Save TTS audio to temp file
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as fp:
        tts.tts_to_file(text=output, file_path=fp.name)
        audio_path = fp.name

    return output, audio_path

# Transcribe user audio answer
def transcribe_audio(audio):
    audio = whisper.load_audio(audio)
    audio = whisper.pad_or_trim(audio)
    mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)
    options = whisper.DecodingOptions()
    result = whisper.decode(whisper_model, mel, options)
    return result.text

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("### Voice Q&A Generator")
    with gr.Row():
        input_text = gr.Textbox(label="Coursebook Text")
        generate_btn = gr.Button("Generate Question")

    question_out = gr.Textbox(label="Generated Question")
    audio_out = gr.Audio(label="AI Question (Audio)", type="filepath")

    with gr.Row():
        user_audio = gr.Audio(source="microphone", type="filepath", label="Your Answer")
        transcribed_text = gr.Textbox(label="Transcribed Answer")

    generate_btn.click(fn=generate_question, inputs=input_text, outputs=[question_out, audio_out])
    user_audio.change(fn=transcribe_audio, inputs=user_audio, outputs=transcribed_text)

demo.launch()