app.py

import gradio as gr
import os
import torchaudio
from speechbrain.pretrained import EncoderClassifier

def accent_detect(video_file):
    # Save uploaded video
    if isinstance(video_file, tuple):
        video_path = video_file[0]
    else:
        video_path = "uploaded_input.mp4"
        with open(video_path, "wb") as f:
            f.write(video_file.read())

    # Extract audio
    os.system(f"ffmpeg -y -i {video_path} -ar 16000 -ac 1 -vn audio.wav")
    if not os.path.exists("audio.wav") or os.path.getsize("audio.wav") < 1000:
        return "Audio extraction failed. Please check your file."

    # Classify accent
    accent_model = EncoderClassifier.from_hparams(
        source="speechbrain/lang-id-commonlanguage_ecapa",
        savedir="tmp_accent_model"
    )
    signal, fs = torchaudio.load("audio.wav")
    if signal.shape[0] > 1:
        signal = signal[0].unsqueeze(0)
    prediction = accent_model.classify_batch(signal)
    pred_label = prediction[3][0]
    pred_scores = prediction[1][0]
    confidence = float(pred_scores.max()) * 100
    explanation = f"Predicted Accent: {pred_label} ({confidence:.1f}%)\nThe model is {confidence:.0f}% confident this is a {pred_label} English accent."
    return explanation

demo = gr.Interface(
    fn=accent_detect,
    inputs=gr.Video(type="filepath", label="Upload a Video File (MP4, WEBM, etc.)"),
    outputs="text",
    title="🗣️ English Accent Classifier (Gradio Demo)",
    description="Upload a short video clip of English speech. This tool predicts the English accent and confidence."
)

if __name__ == "__main__":
    demo.launch()