data/librispeech/utils.py

"""
Copyright (c) Facebook, Inc. and its affiliates.
All rights reserved.

This source code is licensed under the MIT-style license found in the
LICENSE file in the root directory of this source tree.
"""

from __future__ import absolute_import, division, print_function, unicode_literals

import os

import sox


def find_transcript_files(dir):
    files = []
    for dirpath, _, filenames in os.walk(dir):
        for filename in filenames:
            if filename.endswith(".trans.txt"):
                files.append(os.path.join(dirpath, filename))
    return files


def parse_speakers_gender(spk_file):
    ret = {}
    with open(spk_file, "r") as f:
        for line in f:
            if line.startswith(";"):
                continue
            sample_id, gen, _ = line.split("|", 2)
            ret[sample_id.strip()] = gen.strip()
    return ret


def transcript_to_list(file):
    audio_path = os.path.dirname(file)
    ret = []
    with open(file, "r") as f:
        for line in f:
            file_id, trans = line.strip().split(" ", 1)
            audio_file = os.path.abspath(os.path.join(audio_path, file_id + ".flac"))
            duration = sox.file_info.duration(audio_file) * 1000  # miliseconds
            ret.append([file_id, audio_file, str(duration), trans.lower()])

    return ret


def read_list(src, files):
    ret = []
    for file in files:
        with open(os.path.join(src, file + ".lst"), "r") as f:
            for line in f:
                _, _, _, trans = line.strip().split(" ", 3)
                ret.append(trans)

    return ret