audiomnist / audiomnist.py
flexthink
Fixes
00f96ff
# coding=utf-8
# Copyright 2022 Artem Ploujnikov (HuggingFace adaptation only)
#
# Original dataset: https://github.com/soerenab/AudioMNIST
#
# If you use this dataset, please cite the following article:
# @ARTICLE{becker2018interpreting,
# author = {Becker, S\"oren and Ackermann, Marcel and Lapuschkin, Sebastian and M\"uller, Klaus-Robert and Samek, Wojciech},
# title = {Interpreting and Explaining Deep Neural Networks for Classification of Audio Signals},
# journal = {CoRR},
# volume = {abs/1807.03418},
# year = {2018},
# archivePrefix = {arXiv},
# eprint = {1807.03418},
# }
# Lint as: python3
import json
import logging
import re
import os
import datasets
from glob import glob
logger = logging.getLogger(__name__)
_DESCRIPTION = """\
AudioMNIST, a research baseline dataset
"""
_BASE_URL = "https://huggingface.co/datasets/flexthink/audiomnist/resolve/main"
_HOMEPAGE_URL = "https://huggingface.co/datasets/flexthink/audiomnist"
_SPLITS = ["train", "valid", "test"]
_GENDERS = ["female", "male"]
_ACCENTS = [
"Arabic",
"Brasilian",
"Chinese",
"Danish",
"English",
"French",
"German",
"Italian",
"Levant",
"Madras",
"South African",
"South Korean",
"Spanish",
"Tamil",
]
_SAMPLING_RATE = 48000
_ACCENT_MAP = {
"german": "German",
"Egyptian_American?": "Arabic",
"German/Spanish": "German",
}
_META_FILE = "audioMNIST_meta.json"
_RE_FILE_NAME = re.compile("(?P<digit>\d+)_(?P<speaker_id>\d+)_(?P<sample_idx>\d+).wav")
class GraphemeToPhoneme(datasets.GeneratorBasedBuilder):
def __init__(self, base_url=None, splits=None, *args, **kwargs):
super().__init__(*args, **kwargs)
self.base_url = base_url or _BASE_URL
self.splits = splits or _SPLITS
def _info(self):
return datasets.DatasetInfo(
description=_DESCRIPTION,
features=datasets.Features(
{
"file_name": datasets.Value("string"),
"audio": datasets.features.Audio(sampling_rate=_SAMPLING_RATE),
"speaker_id": datasets.Value("string"),
"age": datasets.Value("int8"),
"gender": datasets.ClassLabel(names=_GENDERS),
"accent": datasets.ClassLabel(names=_ACCENTS),
"native_speaker": datasets.Value("bool"),
"origin": datasets.Value("string"),
"digit": datasets.Value("int8"),
},
),
supervised_keys=None,
homepage=_HOMEPAGE_URL,
)
def _get_url(self, split):
return f"{self.base_url}/dataset/{split}.tar.gz"
def _get_meta_url(self):
return f"{self.base_url}/meta/{_META_FILE}"
def _split_generator(self, dl_manager, split):
archive_url = self._get_url(split)
archive_path = dl_manager.download_and_extract(archive_url)
meta_url = self._get_meta_url()
meta_file = dl_manager.download(meta_url)
speaker_map = self._get_speaker_map(meta_file)
return datasets.SplitGenerator(
name=split,
gen_kwargs={
"archive_path": archive_path,
"speaker_map": speaker_map,
},
)
def _get_speaker_map(self, file_name):
with open(file_name) as speaker_file:
result = json.load(speaker_file)
for entry in result.values():
entry["accent"] = _ACCENT_MAP.get(
entry["accent"], entry["accent"])
return result
def _split_generators(self, dl_manager):
return [self._split_generator(dl_manager, split) for split in self.splits]
def _map_speaker_info(self, speaker_info):
result = dict(speaker_info)
result["native_speaker"] = speaker_info["native speaker"] == "yes"
del result["native speaker"]
del result["recordingdate"]
del result["recordingroom"]
return result
def _generate_examples(self, archive_path, speaker_map):
wav_files = glob(os.path.join(archive_path, 'dataset', '**', '*.wav'))
for path in wav_files:
match = _RE_FILE_NAME.search(path)
if not match:
logger.warn(
f"File {path} does not match the naming convention"
)
continue
digit, speaker_id = [
match.group(key) for key in ["digit", "speaker_id"]
]
with open(path, 'rb') as wav_file:
sample = {
"digit": digit,
"speaker_id": speaker_id,
"file_name": os.path.join(archive_path, path),
"audio": {"path": path, "bytes": wav_file.read()},
}
if speaker_id not in speaker_map:
logger.warn(f"Speaker {speaker_id} not found")
speaker_info = speaker_map[speaker_id]
sample.update(self._map_speaker_info(speaker_info))
yield path, sample