app.py

import gradio as gr
import torch
from faster_whisper import WhisperModel
import tempfile
import logging
import os

# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Mapping of model names to Whisper model sizes
MODELS = {
    "Faster Whisper Medium": "Systran/faster-whisper-medium",  # Use the medium model
}

def transcribe_live_audio(audio):
    """Transcribe live audio from the microphone."""
    try:
        # Save the live audio to a temporary file
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_audio:
            temp_audio_path = temp_audio.name
            audio.save(temp_audio_path)  # Save the audio file
        
        # Load the appropriate model
        device = "cuda" if torch.cuda.is_available() else "cpu"
        compute_type = "float32" if device == "cuda" else "int8"
        model = WhisperModel(MODELS["Faster Whisper Medium"], device=device, compute_type=compute_type)
        
        # Transcribe the live audio
        segments, info = model.transcribe(
            temp_audio_path,
            task="transcribe",
            word_timestamps=True,
            repetition_penalty=1.1,
            temperature=[0.0, 0.1, 0.2, 0.3, 0.4, 0.6, 0.8, 1.0],
        )
        transcription = " ".join([segment.text for segment in segments])
        
        # Clean up temporary audio file
        os.remove(temp_audio_path)
        
        return transcription
    except Exception as e:
        logger.error(f"Error in transcribe_live_audio: {str(e)}")
        return f"Error: {str(e)}"

# Define the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# Live Audio Transcription")
    
    gr.Markdown("Transcribe live audio from your microphone using the **Systran/faster-whisper-medium** model.")
    
    # Live audio input
    live_audio_input = gr.Audio(type="filepath", label="Speak into the microphone")
    
    # Transcription output
    live_transcription_output = gr.Textbox(label="Live Transcription")
    
    # Button to start transcription
    live_transcribe_button = gr.Button("Start Transcription")
    
    # Link button to function
    live_transcribe_button.click(
        transcribe_live_audio,
        inputs=[live_audio_input],
        outputs=live_transcription_output
    )

# Launch the Gradio interface
demo.launch()