Audiomate is a library for easy access to audio datasets. It provides the datastructures for accessing/loading different datasets in a generic way. This should ease the use of audio datasets for example for machine learning tasks.
import audiomate
from audiomate.corpus import io
# Download a dataset
esc_downloader = io.ESC50Downloader()
esc_downloader.download('/local/path')
# Load and work with the dataset
esc50 = audiomate.Corpus.load('/local/path', reader='esc-50')
# e.g. Read the audio signal and the label of specific sample/utterance
utterance = esc50.utterances['1-100032-A-0']
samples = utterance.read_samples()
label_list = utterance.label_lists[audiomate.corpus.LL_SOUND_CLASS]
for label in label_list:
print(label.start, label.value)Furthermore it provides tools for interacting with datasets (validation, splitting, subsets, merge, filter), extracting features, feeding samples for training ML models and more.
Currently supported datasets:
- Acoustic Event Dataset
- AudioMNIST
- Mozilla Common Voice
- ESC-50
- Fluent Speech Commands
- Free Spoken Digit Dataset
- German Distant Speech Corpus
- Google Speech Commands
- GTZAN
- LibriSpeech
- M-AILABS Speech Dataset
- MUSAN
- LITIS Rouen Audio scene dataset
- Spoken Wikipedia Corpora
- Tatoeba
- TIMIT
- Urbansound8k
- Voxforge
Currently supported formats:
pip install audiomateInstall the latest development version:
pip install git+https://github.com/ynop/audiomate.gitFor parts of the functionality (e.g. audio format conversion) sox is used. In order to use it, you have to install sox.
# macos
brew install sox
# with support for specific formats
brew install sox --with-lame --with-flac --with-libvorbis
# linux
apt-get install sox
# anaconda for macOS/windows/linux:
conda install -c conda-forge soxIt's recommended to use a virtual environment when developing audiomate. To create one, execute the following command in the project's root directory:
python -m venv .
To install audiomate and all it's dependencies, execute:
pip install -e .
pip install -e .[dev]
pytest
With PyCharm you might have to change the default test runner. Otherwise, it might only suggest to use nose. To do so, go to File > Settings > Tools > Python Integrated Tools (on the Mac it's PyCharm > Preferences > Settings > Tools > Python Integrated Tools) and change the test runner to py.test.
In order to check the runtime of specific parts, pytest-benchmark is used. Benchmarks are normal test functions, but call the benchmark fixture for the code under test.
To run benchmarks:
# Run all
pytest bench
# Specific benchmark
pytest bench/corpus/test_merge_corpus.py
To compare between different runs:
pytest-benchmark compare
The documentation is written in reStructuredText and transformed into various output formats with the help of Sphinx.
To generate the documentation, execute:
pip install -e .[dev]
cd docs
make html
The generated files are written to docs/_build/html.
Versions is handled using bump2version. To bump the version:
bump2version [major,minor,patch,release,num]
In order to directly go to a final relase version (skip .dev/.rc/...):
bump2version [major,minor,patch] --new-version x.x.x
Commands to create a new release on pypi.
rm -rf build
rm -rf dist
python setup.py sdist
python setup.py bdist_wheel
twine upload dist/*