nvidia-nemo--speech/scripts/dataset_processing/get_commonvoice_data.py

# Copyright (c) 2020, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#

# Copyright (c) 2020, SeanNaren.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#

# To convert mp3 files to wav using sox, you must have installed sox with mp3 support
# For example sudo apt-get install libsox-fmt-mp3
import argparse
import csv
import json
import logging
import multiprocessing
import os
import sys
import tarfile
import urllib.request
from multiprocessing.pool import ThreadPool
from pathlib import Path
from typing import List

from tqdm import tqdm

from nemo.utils.tar_utils import safe_extract

parser = argparse.ArgumentParser(description='Downloads and processes Mozilla Common Voice dataset.')
parser.add_argument("--data_root", default='CommonVoice_dataset/', type=str, help="Directory to store the dataset.")
parser.add_argument('--manifest_dir', default='./', type=str, help='Output directory for manifests')
parser.add_argument("--num_workers", default=multiprocessing.cpu_count(), type=int, help="Workers to process dataset.")
parser.add_argument('--sample_rate', default=16000, type=int, help='Sample rate')
parser.add_argument('--n_channels', default=1, type=int, help='Number of channels for output wav files')
parser.add_argument("--log", dest="log", action="store_true", default=False)
parser.add_argument("--cleanup", dest="cleanup", action="store_true", default=False)
parser.add_argument(
    '--files_to_process',
    nargs='+',
    default=['test.tsv', 'dev.tsv', 'train.tsv'],
    type=str,
    help='list of *.csv file names to process',
)
parser.add_argument(
    '--version',
    default='cv-corpus-5.1-2020-06-22',
    type=str,
    help='Version of the dataset (obtainable via https://commonvoice.mozilla.org/en/datasets',
)
parser.add_argument(
    '--language',
    default='en',
    type=str,
    help='Which language to download.(default english,'
    'check https://commonvoice.mozilla.org/en/datasets for more language codes',
)
args = parser.parse_args()
COMMON_VOICE_URL = (
    f"https://voice-prod-bundler-ee1969a6ce8178826482b88e843c335139bd3fb4.s3.amazonaws.com/"
    "{}/{}.tar.gz".format(args.version, args.language)
)
COMMON_VOICE_USER_AGENT = (
    'Mozilla/5.0 (Windows NT 10.0; WOW64) ' 'AppleWebKit/537.36 (KHTML, like Gecko) Chrome/51.0.2704.103 Safari/537.36'
)


def _load_sox():
    try:
        import sox
        from sox import Transformer
    except ImportError:
        raise ImportError(
            "Optional dependency 'sox' is required by this script. Install it with: pip install sox"
        ) from None

    return sox, Transformer


def download_commonvoice_archive(url: str, output_path: str):
    request = urllib.request.Request(url, headers={'User-Agent': COMMON_VOICE_USER_AGENT})
    with urllib.request.urlopen(request) as response, open(output_path, 'wb') as f:
        while True:
            chunk = response.read(1024 * 1024)
            if not chunk:
                break
            f.write(chunk)


def create_manifest(data: List[tuple], output_name: str, manifest_path: str):
    output_file = Path(manifest_path) / output_name
    output_file.parent.mkdir(exist_ok=True, parents=True)

    with output_file.open(mode='w') as f:
        for wav_path, duration, text in tqdm(data, total=len(data)):
            if wav_path != '':
                # skip invalid input files that could not be converted
                f.write(
                    json.dumps({'audio_filepath': os.path.abspath(wav_path), "duration": duration, 'text': text})
                    + '\n'
                )


def process_files(csv_file, data_root, num_workers):
    """Read *.csv file description, convert mp3 to wav, process text.
        Save results to data_root.

    Args:
        csv_file: str, path to *.csv file with data description, usually start from 'cv-'
        data_root: str, path to dir to save results; wav/ dir will be created
    """
    sox, Transformer = _load_sox()
    wav_dir = os.path.join(data_root, 'wav/')
    os.makedirs(wav_dir, exist_ok=True)
    audio_clips_path = os.path.dirname(csv_file) + '/clips/'

    def process(x):
        file_path, text = x
        file_name = os.path.splitext(os.path.basename(file_path))[0]
        text = text.lower().strip()
        audio_path = os.path.join(audio_clips_path, file_path)
        if os.path.getsize(audio_path) == 0:
            logging.warning(f'Skipping empty audio file {audio_path}')
            return '', '', ''

        output_wav_path = os.path.join(wav_dir, file_name + '.wav')

        if not os.path.exists(output_wav_path):
            tfm = Transformer()
            tfm.rate(samplerate=args.sample_rate)
            tfm.channels(n_channels=args.n_channels)
            tfm.build(input_filepath=audio_path, output_filepath=output_wav_path)

        duration = sox.file_info.duration(output_wav_path)
        return output_wav_path, duration, text

    logging.info('Converting mp3 to wav for {}.'.format(csv_file))
    with open(csv_file) as csvfile:
        reader = csv.DictReader(csvfile, delimiter='\t')
        next(reader, None)  # skip the headers
        data = []
        for row in reader:
            file_name = row['path']
            # add the mp3 extension if the tsv entry does not have it
            if not file_name.endswith('.mp3'):
                file_name += '.mp3'
            data.append((file_name, row['sentence']))
        with ThreadPool(num_workers) as pool:
            data = list(tqdm(pool.imap(process, data), total=len(data)))
    return data


def main():
    if args.log:
        logging.basicConfig(level=logging.INFO)

    data_root = args.data_root
    os.makedirs(data_root, exist_ok=True)

    target_unpacked_dir = os.path.join(data_root, "CV_unpacked")

    if os.path.exists(target_unpacked_dir):
        logging.info('Find existing folder {}'.format(target_unpacked_dir))
    else:
        logging.info("Could not find Common Voice, Downloading corpus...")

        # some dataset versions are packaged in different named files, so forcing
        output_archive_filename = args.language + '.tar.gz'
        output_archive_filename = os.path.join(data_root, output_archive_filename)

        download_commonvoice_archive(COMMON_VOICE_URL, output_archive_filename)
        filename = f"{args.language}.tar.gz"
        target_file = os.path.join(data_root, os.path.basename(filename))

        os.makedirs(target_unpacked_dir, exist_ok=True)
        logging.info("Unpacking corpus to {} ...".format(target_unpacked_dir))
        with tarfile.open(target_file) as tar:
            safe_extract(tar, target_unpacked_dir)
        if args.cleanup:
            logging.info("removing tar archive to save space")
            os.remove(target_file)

    folder_path = os.path.join(target_unpacked_dir, args.version + f'/{args.language}/')
    if not os.path.isdir(folder_path):
        # try without language
        folder_path = os.path.join(target_unpacked_dir, args.version)
        if not os.path.isdir(folder_path):
            # try without version
            folder_path = target_unpacked_dir
            if not os.path.isdir(folder_path):
                logging.error(f'unable to locate unpacked files in {folder_path}')
                sys.exit()

    for csv_file in args.files_to_process:
        data = process_files(
            csv_file=os.path.join(folder_path, csv_file),
            data_root=os.path.join(data_root, os.path.splitext(csv_file)[0]),
            num_workers=args.num_workers,
        )
        logging.info('Creating manifests...')
        create_manifest(
            data=data,
            output_name=f'commonvoice_{os.path.splitext(csv_file)[0]}_manifest.json',
            manifest_path=args.manifest_dir,
        )


if __name__ == "__main__":
    main()