jarvis-server-v2/jarvis/utils/faster_whisper_utils.py

import logging

from faster_whisper import WhisperModel


def load_model():
    log_level = logging.getLogger().level
    global model
    model = WhisperModel('small', device="cpu", cpu_threads=8, compute_type="int8")
    logging.getLogger().setLevel(log_level)


def get_model():
    return model


def faster_whisper_stt(audio_file):
    """
    Transcribe audio file using faster_whisper, no additional server/service needed, runs on CPU.

    :param audio_file:
    :param model:
    :return: text
    """
    if model is None:
        logging.error("Model is not loaded")
        load_model()

    segments, info = model.transcribe(audio_file, beam_size=5)
    print("Detected language '%s' with probability %f" % (info.language, info.language_probability))

    # combines all segments in one string
    text = ''
    for segment in segments:
        text += segment.text + ' '

    return text