app.py

import gradio as gr
from transformers import pipeline
import numpy as np

# Load the model
model_id = "badrex/mms-300m-arabic-dialect-identifier"  # Replace with your model ID
classifier = pipeline("audio-classification", model=model_id)

# Define dialect names for better display
dialect_mapping = {
    "MSA": "Modern Standard Arabic",
    "Egyptian": "Egyptian Arabic",
    "Gulf": "Gulf Arabic",
    "Levantine": "Levantine Arabic",
    "Maghrebi": "Maghrebi Arabic"
}

def predict_dialect(audio):
    # The audio input from Gradio is a tuple of (sample_rate, audio_array)
    if isinstance(audio, tuple) and len(audio) == 2:
        sr, audio_array = audio
    else:
        # Handle error case
        return {"Error": 1.0}
    
    # Process the audio input
    if len(audio_array.shape) > 1:
        audio_array = audio_array.mean(axis=1)  # Convert stereo to mono
    
    # Classify the dialect
    predictions = classifier({"sampling_rate": sr, "raw": audio_array})
    
    # Format results for display
    results = {}
    for pred in predictions:
        dialect_name = dialect_mapping.get(pred['label'], pred['label'])
        results[dialect_name] = float(pred['score'])
    
    return results

# Create the Gradio interface
demo = gr.Interface(
    fn=predict_dialect,
    inputs=gr.Audio(),  # Simplified audio input
    outputs=gr.Label(num_top_classes=5, label="Predicted Dialect"),
    title="Arabic Dialect Identifier",
    description="""This demo identifies Arabic dialects from speech audio.
    Upload an audio file or record your voice speaking Arabic to see which dialect it matches.
    The model identifies: Modern Standard Arabic (MSA), Egyptian, Gulf, Levantine, and Maghrebi dialects.""",
    examples=[
        # Optional: Add example audio files here if you have them
        # ["examples/msa_example.wav"],
        # ["examples/egyptian_example.wav"],
    ],
    allow_flagging="never"
)

# Launch the app
demo.launch()