import gradio as gr
from moviepy.editor import VideoFileClip
from speechbrain.pretrained import EncoderClassifier
import torchaudio
from pytubefix import YouTube
from pytubefix.cli import on_progress
import requests
import os
import torch

CLASSIFIER = "Jzuluaga/accent-id-commonaccent_xlsr-en-english"

def get_default_device():
    """Return the default device (cuda if available, else cpu)."""
    return torch.device("cuda" if torch.cuda.is_available() else "cpu")

def download_video(url):
    """Download video from YouTube or direct MP4 URL using pytubefix."""
    try:
        if "youtube.com" in url or "youtu.be" in url:
            yt = YouTube(url, on_progress_callback=on_progress)
            stream = yt.streams.filter(progressive=True, file_extension='mp4').first()
            if not stream:
                raise ValueError("No suitable video stream found.")
            video_path = stream.download()
            return video_path
        else:
            # Direct MP4 file download
            local_filename = "temp_video.mp4"
            with requests.get(url, stream=True) as r:
                r.raise_for_status()
                with open(local_filename, 'wb') as f:
                    for chunk in r.iter_content(chunk_size=8192):
                        f.write(chunk)
            return local_filename
    except Exception as e:
        raise RuntimeError(f"Failed to download video: {e}")

def extract_audio(video_path):
    """Extract audio from video and save as WAV file."""
    clip = VideoFileClip(video_path)
    audio_path = "temp_audio.wav"
    clip.audio.write_audiofile(audio_path, logger=None)
    clip.close()
    return audio_path

def classify_accent(audio_path):
    """Classify English accent from audio file using SpeechBrain model."""
    device = get_default_device()
    classifier = EncoderClassifier.from_hparams(
        source=CLASSIFIER,
        savedir="pretrained_models/accent_classifier",
        run_opts={"device": str(device)}
    )
    waveform, sample_rate = torchaudio.load(audio_path)
    prediction = classifier.classify_batch(waveform.to(device))
    predicted_accent = prediction[3][0]
    confidence = prediction[1].exp().max().item() * 100
    return predicted_accent, f"{confidence:.2f}%"

def process_video(url):
    """Main processing pipeline: download video, extract audio, classify accent."""
    video_path = None
    audio_path = None
    try:
        video_path = download_video(url)
        audio_path = extract_audio(video_path)
        accent, confidence = classify_accent(audio_path)
        return accent, confidence
    except Exception as e:
        return f"Error: {e}", ""
    finally:
        # Clean up temporary files
        for f in [video_path, audio_path]:
            if f and os.path.exists(f):
                os.remove(f)

iface = gr.Interface(
    fn=process_video,
    inputs=gr.Textbox(label="Enter Public Video URL (YouTube, Loom, direct MP4)"),
    outputs=[
        gr.Textbox(label="Detected Accent"),
        gr.Textbox(label="Confidence Score")
    ],
    title="English Accent Classifier",
    description="Paste a public video URL to detect the English accent and confidence score."
)

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