"""faster-whisper large-v3 transcription with VAD."""

from functools import lru_cache

import torch
from faster_whisper import WhisperModel

WHISPER_MODEL = "large-v3"


@lru_cache(maxsize=1)
def get_whisper_model():
    device = "cuda" if torch.cuda.is_available() else "cpu"
    compute_type = "float16" if device == "cuda" else "int8"
    return WhisperModel(WHISPER_MODEL, device=device, compute_type=compute_type)


def transcribe(audio_path, language=None):
    """Transcribe audio. Returns list of {start, end, text}."""
    model = get_whisper_model()
    segs, _info = model.transcribe(
        audio_path,
        language=language,
        beam_size=5,
        vad_filter=True,
        vad_parameters=dict(min_silence_duration_ms=500),
    )
    return [{"start": s.start, "end": s.end, "text": s.text.strip()} for s in segs]