import gradio as gr import os import torchaudio from speechbrain.pretrained import EncoderClassifier def accent_detect(video_link, video_file): # Decide which input to use video_path = None # If a video file is uploaded, use it if video_file is not None: video_path = "uploaded_input.mp4" with open(video_path, "wb") as f: f.write(video_file.read()) # Else if a link is provided, try to download it elif video_link and len(video_link.strip()) > 8: # Use yt-dlp for YouTube or wget for direct link if "youtube.com" in video_link or "youtu.be" in video_link: os.system(f'yt-dlp -o input_video.mp4 "{video_link}"') else: os.system(f'wget -O input_video.mp4 "{video_link}"') if os.path.exists("input_video.mp4") and os.path.getsize("input_video.mp4") > 0: video_path = "input_video.mp4" else: return "Failed to download the video. Please check your link." else: return "Please upload a video file or provide a valid video link." # Extract audio from video 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 use a different video." # Load model and 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}%)\n" f"The model is {confidence:.0f}% confident this is a {pred_label} English accent." ) return explanation demo = gr.Interface( fn=accent_detect, inputs=[ gr.Textbox(label="YouTube or direct MP4 link (optional)", placeholder="https://youtube.com/yourvideo"), gr.File(label="Or upload a video file (MP4, WEBM, etc.)"), ], outputs="text", title="🗣️ English Accent Classifier (Gradio Demo)", description="Paste a YouTube/direct MP4 link or upload a video file with English speech. The tool predicts the English accent and confidence." ) if __name__ == "__main__": demo.launch()