app.py

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

print("=== Application Starting ===")

# define dialect mapping
dialect_mapping = {
    "MSA": "Modern Standard Arabic (MSA) - العربية الفصحى الحديثة",
    "Egyptian": "Egyptian Arabic -  اللهجة المصرية العامية", 
    "Gulf": "Peninsular Arabic - لهجة الجزيرة العربية",
    "Levantine": "Levantine Arabic - لهجة بلاد الشام",
    "Maghrebi": "Maghrebi Arabic - اللهجة المغاربية"
}

@spaces.GPU
def predict_dialect(audio):
    # load model inside the GPU function
    print("Loading model on GPU...")
    model_id = "badrex/mms-300m-arabic-dialect-identifier"
    classifier = pipeline("audio-classification", model=model_id)  # no device specified
    print("Model loaded successfully")
    
    if audio is None:
        return {"Error": 1.0}
    
    sr, audio_array = audio
    
    if len(audio_array.shape) > 1:
        audio_array = audio_array.mean(axis=1)

    if audio_array.dtype != np.float32:
        if audio_array.dtype == np.int16:
            audio_array = audio_array.astype(np.float32) / 32768.0
        else:
            audio_array = audio_array.astype(np.float32)
    
    print(f"Processing audio: sample rate={sr}, shape={audio_array.shape}")
    
    # classify the dialect
    predictions = classifier({"sampling_rate": sr, "raw": audio_array})
    
    # format results
    results = {}
    for pred in predictions:
        dialect_name = dialect_mapping.get(pred['label'], pred['label'])
        results[dialect_name] = float(pred['score'])
    
    return results

# prepare examples
examples = []
examples_dir = "examples"
if os.path.exists(examples_dir):
    for filename in os.listdir(examples_dir):
        if filename.endswith((".wav", ".mp3", ".ogg")):
            examples.append([os.path.join(examples_dir, filename)])
    print(f"Found {len(examples)} example files")

description = """
By <a href="https://badrex.github.io/">Badr Alabsi</a> with ❤️🤍💚

This demo uses a Transformer-based model for Spoken Arabic Dialect Identification. 
Upload an audio file or record yourself speaking to identify the Arabic dialect!
"""

demo = gr.Interface(
    fn=predict_dialect,
    inputs=gr.Audio(),
    outputs=gr.Label(num_top_classes=5, label="Predicted Dialect"),
    title="Tamyïz 🍉 Arabic Dialect Identification in Speech",
    description=description,
    examples=examples if examples else None,
    cache_examples=False,
    flagging_mode=None
)

print("=== Launching demo ===")
demo.launch()