Transcriptor/core/transcribe.py

import time
import faster_whisper
from faster_whisper.transcribe import Segment


def transcribe(
    file: str,
    model_name: str = "large-v3",
    device: str = "cuda",
    language: str = "es",
    download_root: str = "/home/superti/workspace/models",
    initial_prompt: str = "",
) -> tuple[list[Segment], float]:
    whisper = faster_whisper.WhisperModel(model_name, device=device, download_root=download_root)
    start = time.time()
    trns, _ = whisper.transcribe(
        file,
        word_timestamps=True,
        language=language,
        initial_prompt=initial_prompt or None,
    )
    segments = list(trns)
    elapsed = time.time() - start
    return segments, elapsed