examples/wenet/toolbox_infer.py

#!/usr/bin/python3
# -*- coding: utf-8 -*-
import argparse
import os
from pathlib import Path
import sys
import tempfile

pwd = os.path.abspath(os.path.dirname(__file__))
sys.path.append(os.path.join(pwd, "../../"))

import librosa
import numpy as np
import sherpa
from scipy.io import wavfile
import torch
import torchaudio

from project_settings import project_path, temp_directory
from toolbox.k2_sherpa.utils import audio_convert
from toolbox.k2_sherpa import decode, nn_models


def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--model_dir",
        default=(project_path / "pretrained_models/huggingface/csukuangfj/wenet-chinese-model").as_posix(),
        type=str
    )
    parser.add_argument(
        "--in_filename",
        default=(project_path / "data/test_wavs/paraformer-zh/si_chuan_hua.wav").as_posix(),
        type=str
    )
    parser.add_argument("--sample_rate", default=16000, type=int)
    args = parser.parse_args()
    return args


def main():
    args = get_args()

    # audio convert
    in_filename = Path(args.in_filename)
    out_filename = Path(tempfile.gettempdir()) / "asr" / in_filename.name
    out_filename.parent.mkdir(parents=True, exist_ok=True)

    audio_convert(in_filename=in_filename.as_posix(),
                  out_filename=out_filename.as_posix(),
                  )

    # load recognizer
    m_dict = nn_models.model_map["Chinese"][0]

    local_model_dir = Path(args.model_dir)
    nn_model_file = local_model_dir / m_dict["nn_model_file"]
    tokens_file = local_model_dir / m_dict["tokens_file"]

    recognizer = nn_models.load_recognizer(
        repo_id=m_dict["repo_id"],
        nn_model_file=nn_model_file.as_posix(),
        tokens_file=tokens_file.as_posix(),
        sub_folder=m_dict["sub_folder"],
        local_model_dir=local_model_dir,
        recognizer_type=m_dict["recognizer_type"],
        decoding_method="greedy_search",
        num_active_paths=2,
    )

    text = decode.decode_by_recognizer(recognizer=recognizer,
                                       filename=out_filename.as_posix(),
                                       )
    print("text: {}".format(text))
    return


if __name__ == "__main__":
    main()