import ffmpeg
import os
from faster_whisper import WhisperModel


def extract_audio(video_path: str) -> str:
    """Extract mono 16kHz WAV audio from a video file."""
    audio_path = os.path.splitext(video_path)[0] + "_audio.wav"
    (
        ffmpeg
        .input(video_path)
        .output(audio_path, acodec="pcm_s16le", ac=1, ar="16000")
        .overwrite_output()
        .run(quiet=True)
    )
    return audio_path


def transcribe(video_path: str, model_size: str = "base") -> dict:
    """
    Transcribe video audio using faster-whisper.
    Returns a dict with text, segments (with word timestamps), and language.
    """
    audio_path = extract_audio(video_path)

    # CPU with int8 quantization — fast, no GPU needed
    model = WhisperModel(model_size, device="cpu", compute_type="int8")

    segments_iter, info = model.transcribe(
        audio_path,
        word_timestamps=True,
        vad_filter=True,  # skip silence automatically
    )

    # Convert to standard format
    segments = []
    full_text = []

    for seg in segments_iter:
        words = []
        if seg.words:
            for w in seg.words:
                words.append({
                    "word": w.word,
                    "start": w.start,
                    "end": w.end,
                    "probability": w.probability,
                })
        segments.append({
            "id": seg.id,
            "start": seg.start,
            "end": seg.end,
            "text": seg.text.strip(),
            "words": words,
        })
        full_text.append(seg.text.strip())

    # Clean up audio file to save disk space
    try:
        os.remove(audio_path)
    except Exception:
        pass

    return {
        "text": " ".join(full_text),
        "segments": segments,
        "language": info.language,
    }