app.py

import os
import spaces
import torch
import torchaudio
import gradio as gr
import logging
from whosper import WhosperTranscriber


logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)


if torch.cuda.is_available():
    device = "cuda"
    logger.info("Using CUDA for inference.")
elif torch.backends.mps.is_available():
    device = "mps"
    logger.info("Using MPS for inference.")
else:
    device = "cpu"
    logger.info("Using CPU for inference.")


model_id = "sudoping01/maliba-asr-v1"
transcriber = WhosperTranscriber(model_id=model_id)
logger.info(f"Transcriber initialized with model: {model_id}")

def resample_audio(audio_path, target_sample_rate=16000):

    """
    Converts the audio file to the target sampling rate (16000 Hz).
    
    Args:
        audio_path (str): Path to the audio file.
        target_sample_rate (int): The desired sample rate.
    Returns:
        A tensor containing the resampled audio data and the target sample rate.
    """
    try:
        waveform, original_sample_rate = torchaudio.load(audio_path)
        
        if original_sample_rate != target_sample_rate:
            resampler = torchaudio.transforms.Resample(
                orig_freq=original_sample_rate, 
                new_freq=target_sample_rate
            )
            waveform = resampler(waveform)
        
        return waveform, target_sample_rate
    except Exception as e:
        logger.error(f"Error resampling audio: {e}")
        raise e

@spaces.GPU()
def transcribe_audio(audio_file):

    """
    Transcribes the provided audio file into Bambara text using Whosper.
    
    Args:
        audio_file: The path to the audio file to transcribe.
    Returns:
        A string representing the transcribed Bambara text.
    """

    if audio_file is None:
        return "Please provide an audio file for transcription."
    
    try:
        logger.info(f"Transcribing audio file: {audio_file}")
        

        result = transcriber.transcribe_audio(audio_file)
        
        logger.info("Transcription successful.")
        return result.get("text", "")
        
    except Exception as e:
        logger.error(f"Transcription failed: {e}")
        return f"Error during transcription: {str(e)}"

def get_example_files(directory="./examples"):

    """
    Returns a list of audio files from the examples directory.
    
    Args:
        directory (str): The directory to search for audio files.
    Returns:
        list: A list of paths to the audio files.
    """

    if not os.path.exists(directory):
        logger.warning(f"Examples directory {directory} not found.")
        return []
    

    audio_extensions = ['.wav', '.mp3', '.m4a', '.flac', '.ogg']
    audio_files = []
    
    try:
        files = os.listdir(directory)
        for file in files:
            if any(file.lower().endswith(ext) for ext in audio_extensions):
                full_path = os.path.abspath(os.path.join(directory, file))
                audio_files.append(full_path)
        
        logger.info(f"Found {len(audio_files)} example audio files.")
        return audio_files[:5]  
        
    except Exception as e:
        logger.error(f"Error reading examples directory: {e}")
        return []

def build_interface():
    """
    Builds the Gradio interface for Bambara speech recognition.
    """

    example_files = get_example_files()
    
    with gr.Blocks(title="Bambara Speech Recognition") as demo:
        gr.Markdown(
            """
            # 🎤 Bambara Automatic Speech Recognition
            
            **Powered by MALIBA-AI**
            
            Convert Bambara speech to text using our state-of-the-art ASR model. You can either:
            - 🎙️ **Record** your voice directly
            - 📁 **Upload** an audio file
            - 🎵 **Try** our example audio files
            
            ## Supported Audio Formats
            WAV, MP3, M4A, FLAC, OGG
            """
        )
        
        with gr.Row():
            with gr.Column():

                audio_input = gr.Audio(
                    label="🎤 Record or Upload Audio",
                    type="filepath",
                    sources=["microphone", "upload"]
                )
                
                transcribe_btn = gr.Button(
                    "🔄 Transcribe Audio", 
                    variant="primary",
                    size="lg"
                )
                

                clear_btn = gr.Button("🗑️ Clear", variant="secondary")
            
            with gr.Column():
                output_text = gr.Textbox(
                    label="📝 Transcribed Text (Bambara)",
                    lines=8,
                    placeholder="Your transcribed Bambara text will appear here...",
                    interactive=False
                )
        

        if example_files:
            gr.Markdown("## 🎵 Try These Examples")
            gr.Examples(
                examples=[[f] for f in example_files],
                inputs=[audio_input],
                outputs=output_text,
                fn=transcribe_audio,
                cache_examples=False,
                label="Example Audio Files"
            )
        

        gr.Markdown(
            """
            ---
            
            ## ℹ️ About This Model
            
            - **Model:** [sudoping01/maliba-asr-v1](https://huggingface.co/sudoping01/maliba-asr-v1)
            - **Developer:** MALIBA-AI
            - **Language:** Bambara (bm)
            - **Task:** Automatic Speech Recognition (ASR)
            - **Sample Rate:** 16kHz (automatically resampled)
            
            ## 🚀 How to Use
            
            1. **Record Audio:** Click the microphone button and speak in Bambara
            2. **Upload File:** Click the upload button to select an audio file
            3. **Transcribe:** Click the "Transcribe Audio" button
            4. **View Results:** See your transcribed text in Bambara
            
            ## 📊 Performance Notes
            
            - Best results with clear speech and minimal background noise
            - Supports various audio formats and durations
            - Optimized for Bambara language patterns and phonetics
            """
        )
        

        transcribe_btn.click(
            fn=transcribe_audio,
            inputs=[audio_input],
            outputs=output_text,
            show_progress=True
        )
        
        clear_btn.click(
            fn=lambda: (None, ""),
            outputs=[audio_input, output_text]
        )
        

        audio_input.change(
            fn=transcribe_audio,
            inputs=[audio_input],
            outputs=output_text,
            show_progress=True
        )
    
    return demo

def main():
    """
    Main function to launch the Gradio interface.
    """
    logger.info("Starting Bambara ASR Gradio interface.")
    

    interface = build_interface()
    interface.launch(
        share=False,
        server_name="0.0.0.0",
        server_port=7860
    )
    
    logger.info("Gradio interface launched successfully.")

if __name__ == "__main__":
    main()