whispercpp transcription using python bindings

whispercpp.py

import ffmpeg
import numpy as np
import functools as f
from whispercpp import Whisper
from pathlib import Path


def convert_to_numpy_array(file_path: str | Path):
    try:
        out, _ = (
            ffmpeg.input(file_path, threads=0)
            .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar="16k")
            .overwrite_output()
            .run(capture_stdout=True, capture_stderr=True)
        )

        return np.frombuffer(out, np.int16).flatten().astype(np.float32) / 32768.0
    except ffmpeg.Error as e:
        raise RuntimeError(f"Failed to load audio: {e.stderr.decode}")


@f.lru_cache(maxsize=1)
def get_whisper() -> Whisper:
    whisper_instance = Whisper.from_pretrained("base")
    whisper_instance.params.with_language("pt")
    return whisper_instance


def transcribe_audio(file_path: str | Path):
    file_data = convert_to_numpy_array(file_path)
    whisper = get_whisper()
    return whisper.transcribe(file_data)