需要哪些测试覆盖方法？
依赖项：版本控制 (检查 commit ID 日期)
参见: requirements.txt
运行:find *txt . | grep extraDocstring 测试：注释掉的函数签名
(针对外部调用的函数)按功能关键代码块的单元测试
针对普通实验的集成测试，以涵盖通用任务基础上的用例
回归测试：稳定接口及其重构
用于自动检查和修正 Python 和 YAML 代码风格的 Linter
如何参与贡献？
提出您的问题并在Discussion 区参与讨论
报告错误或请求功能，在Issue 区提出
通过 SpeechBrain 发布预训练模型
例如，将 HuggingFace 账户链接到 SpeechBrain 以托管您的模型卡
GitHub 工作流程：基于配置的策略
API 配置位于.github/workflows
(所有配置都创建一个一次性的 ubuntu-latest 环境)
信息：尽管我们的 PyTorch 要求是
torch>=1.9.0
torchaudio>=0.9.0
我们的测试仅覆盖一个 PyTorch 版本，即最新版本。
pre-commit.yml
SpeechBrain pre-commit / pre-commit (pull_request)
python-version: ‘3.8’
运行 pre-commit action，配置在.pre-commit-config.yaml 中
hook: https://github.com/pre-commit/pre-commit-hooks
trailing-whitespace
end-of-file-fixer
requirements-txt-fixer
mixed-line-ending
check-added-large-fileshook: https://github.com/psf/black
black
clickhook: https://gitlab.com/pycqa/flake8.git
flake8; 参见: .flake8hook: https://adrienverge.net/yamllint/
yamllint; 参见: .yamllint.yaml
pythonapp.yml
SpeechBrain toolkit CI / Tests (3.7) (pull_request)
SpeechBrain toolkit CI / Tests (3.8) (pull_request)
SpeechBrain toolkit CI / Tests (3.9) (pull_request)
python-version: [3.7, 3.8, 3.9]
创建全新的环境
sudo apt-get install -y libsndfile1 pip install -r requirements.txt pip install --editable . pip install ctc-segmentation
运行 PyTest 检查
参见: pytest.ini - 文件:test_*.py;check_*.py;example_*.py及 norecursedirs
参见: conftest.py - 准备测试项收集和直接发现# excerpts parser.addoption("--device", action="store", default="cpu") ... try: import numba # noqa: F401 except ModuleNotFoundError: collect_ignore.append("speechbrain/nnet/loss/transducer_loss.py") ...
a. hook: 使用 pytest 进行一致性测试
pytest tests/consistencyb. hook: 使用 pytest 进行单元测试
pytest tests/unittestsc. hook: 使用 pytest 进行 Doctest
pytest --doctest-modules speechbraind. hook: 使用 pytest 进行集成测试
pytest tests/integration
verify-docs-gen.yml [I.2.a]
验证文档生成 / docs (pull_request)
python-version: ‘3.8’
创建全新的环境
pip install -r requirements.txt pip install --editable . pip install -r docs/docs-requirements.txt
生成文档
cd docs make html
对比: .readthedocs.yaml - python version: 3.8
newtag.yml
在推送新 tag 时起草版本发布
对
develop分支的 commit ID 进行 tagging之前
完成 tests/PRE-RELEASE-TESTS.md 中的测试
设置全新的环境
运行
pytesta. hook: tests/.run-load-yaml-tests.sh
b. hook: tests/.run-recipe-tests.sh
c. hook: tests/.run-HF-checks.sh
d. hook: tests/.run-url-checks.sh
将 speechbrain/version.txt 更新到下一个版本
action: 起草推送到
main分支
意味着 pre-push hook，参见: .pre-push-config.yaml 中的 hook
release.yml
发布到 PyPI
python-version: 3.8
action: checkout 到
main分支creates:
pypa/build用于构建二进制 wheel 和源 tarballaction: 通过
pypa/gh-action-pypi-publish@master发布到 PyPI
意味着使用pyproject.toml - target-version = [’py38’]
-
python_requires=”>=3.7”,
uses: speechbrain/version.txt
需要
"hyperpyyaml", "joblib", "numpy", "packaging", "scipy", "sentencepiece", "torch>=1.9", "torchaudio", "tqdm", "huggingface_hub",
指向 https://speechbrain.cn/
这些检查中使用的工具/hook 的版本通过 lint-requirements.txt 控制，它是 requirements.txt 中的一个嵌套依赖。随着 SpeechBrain 主要版本的发布，每个 hook 的版本应随之更新——同时确保源文件、测试和构建（包括运行拼写检查）中的依赖一致性。
注意：PyTorch 关于 Python 版本的说明 (截至 2022-11-09)
建议您使用 Python 3.6、3.7 或 3.8
用于报告代码覆盖率的 PyTest
如何了解待合并 Open PR 的测试覆盖率变化？
(仅 CPU 片段)
# Example: install more dependencies to avoid ignoring modules
sudo apt install -y libsndfile1
pip install ctc_segmentation
# install coverage
pip install pytest-cov
# run the test (w/ duration reporting)
pytest --durations=0 --cov=speechbrain --cov-context=test --doctest-modules speechbrain tests --ignore=speechbrain/nnet/loss/transducer_loss.py
示例：收集 459 个测试项后，报告“缺失”语句 4481/16782 (覆盖率 73%)。
然而——核心模块的 Python 代码并非全部需要覆盖；到目前为止，仅保证了一致性。
延伸阅读
pytest 与 coverage - https://breadcrumbscollector.tech/how-to-use-code-coverage-in-python-with-pytest/ (@Adel-Moumen 推荐)
pytest --durations=0 --cov=speechbrain --cov-context=test --doctest-modules speechbrain tests --ignore=speechbrain/nnet/loss/transducer_loss.py
---------- coverage: platform linux, python 3.9.12-final-0 -----------
Name Stmts Miss Cover
-----------------------------------------------------------------------------
speechbrain/alignment/aligner.py 380 61 84%
speechbrain/alignment/ctc_segmentation.py 189 10 95%
speechbrain/core.py 424 155 63% <== < 80%
speechbrain/dataio/batch.py 99 8 92%
speechbrain/dataio/dataio.py 279 50 82%
speechbrain/dataio/dataloader.py 140 25 82%
speechbrain/dataio/dataset.py 100 8 92%
speechbrain/dataio/encoder.py 328 46 86%
speechbrain/dataio/iterators.py 80 62 22% <== < 80%
speechbrain/dataio/legacy.py 121 41 66% <== < 80%
speechbrain/dataio/preprocess.py 22 4 82%
speechbrain/dataio/sampler.py 224 61 73% <== < 80%
speechbrain/dataio/wer.py 63 54 14% <== < 80%
speechbrain/decoders/ctc.py 111 89 20% <== < 80%
speechbrain/decoders/seq2seq.py 370 46 88%
speechbrain/decoders/transducer.py 133 64 52% <== < 80%
speechbrain/lm/arpa.py 77 3 96%
speechbrain/lm/counting.py 37 4 89%
speechbrain/lm/ngram.py 36 1 97%
speechbrain/lobes/augment.py 154 55 64% <== < 80%
speechbrain/lobes/beamform_multimic.py 20 14 30% <== < 80%
speechbrain/lobes/features.py 96 9 91%
speechbrain/lobes/models/CRDNN.py 52 12 77% <== < 80%
speechbrain/lobes/models/ContextNet.py 83 3 96%
speechbrain/lobes/models/ECAPA_TDNN.py 157 7 96%
speechbrain/lobes/models/HifiGAN.py 321 146 55% <== < 80%
speechbrain/lobes/models/MetricGAN.py 74 29 61% <== < 80%
speechbrain/lobes/models/Tacotron2.py 364 66 82%
speechbrain/lobes/models/conv_tasnet.py 121 6 95%
speechbrain/lobes/models/dual_path.py 357 55 85%
speechbrain/lobes/models/fairseq_wav2vec.py 93 93 0% <== < 80%
speechbrain/lobes/models/g2p/dataio.py 136 107 21% <== < 80%
speechbrain/lobes/models/g2p/homograph.py 118 20 83%
speechbrain/lobes/models/g2p/model.py 132 109 17% <== < 80%
speechbrain/lobes/models/huggingface_wav2vec.py 145 47 68% <== < 80%
speechbrain/lobes/models/resepformer.py 180 21 88%
speechbrain/lobes/models/segan_model.py 102 88 14% <== < 80%
speechbrain/lobes/models/transformer/Conformer.py 111 7 94%
speechbrain/lobes/models/transformer/Transformer.py 180 22 88%
speechbrain/lobes/models/transformer/TransformerASR.py 92 28 70% <== < 80%
speechbrain/lobes/models/transformer/TransformerLM.py 47 5 89%
speechbrain/lobes/models/transformer/TransformerSE.py 20 2 90%
speechbrain/lobes/models/transformer/TransformerST.py 81 60 26% <== < 80%
speechbrain/lobes/models/wav2vec.py 123 55 55% <== < 80%
speechbrain/nnet/CNN.py 417 56 87%
speechbrain/nnet/RNN.py 471 51 89%
speechbrain/nnet/activations.py 39 1 97%
speechbrain/nnet/attention.py 234 44 81%
speechbrain/nnet/complex_networks/c_CNN.py 130 23 82%
speechbrain/nnet/complex_networks/c_RNN.py 374 67 82%
speechbrain/nnet/complex_networks/c_normalization.py 277 68 75% <== < 80%
speechbrain/nnet/complex_networks/c_ops.py 108 40 63% <== < 80%
speechbrain/nnet/containers.py 139 14 90%
speechbrain/nnet/linear.py 27 1 96%
speechbrain/nnet/loss/si_snr_loss.py 20 16 20% <== < 80%
speechbrain/nnet/loss/stoi_loss.py 81 1 99%
speechbrain/nnet/loss/transducer_loss.py 136 136 0% <== < 80%
speechbrain/nnet/losses.py 323 112 65% <== < 80%
speechbrain/nnet/normalization.py 142 6 96%
speechbrain/nnet/pooling.py 156 31 80%
speechbrain/nnet/quantisers.py 47 2 96%
speechbrain/nnet/quaternion_networks/q_CNN.py 150 25 83%
speechbrain/nnet/quaternion_networks/q_RNN.py 370 59 84%
speechbrain/nnet/quaternion_networks/q_linear.py 50 11 78% <== < 80%
speechbrain/nnet/quaternion_networks/q_normalization.py 44 4 91%
speechbrain/nnet/quaternion_networks/q_ops.py 229 122 47% <== < 80%
speechbrain/nnet/schedulers.py 363 103 72% <== < 80%
speechbrain/nnet/transducer/transducer_joint.py 33 5 85%
speechbrain/pretrained/fetching.py 48 6 88%
speechbrain/pretrained/interfaces.py 786 338 57% <== < 80%
speechbrain/pretrained/training.py 33 28 15% <== < 80%
speechbrain/processing/PLDA_LDA.py 345 96 72% <== < 80%
speechbrain/processing/decomposition.py 102 8 92%
speechbrain/processing/diarization.py 319 157 51% <== < 80%
speechbrain/processing/features.py 359 75 79% <== < 80%
speechbrain/processing/multi_mic.py 345 2 99%
speechbrain/processing/signal_processing.py 166 39 77% <== < 80%
speechbrain/processing/speech_augmentation.py 386 34 91%
speechbrain/tokenizers/SentencePiece.py 181 74 59% <== < 80%
speechbrain/utils/Accuracy.py 24 17 29% <== < 80%
speechbrain/utils/DER.py 44 33 25% <== < 80%
speechbrain/utils/bleu.py 50 43 14% <== < 80%
speechbrain/utils/callchains.py 28 5 82%
speechbrain/utils/checkpoints.py 294 52 82%
speechbrain/utils/data_pipeline.py 181 15 92%
speechbrain/utils/data_utils.py 197 77 61% <== < 80%
speechbrain/utils/depgraph.py 82 1 99%
speechbrain/utils/distributed.py 61 37 39% <== < 80%
speechbrain/utils/edit_distance.py 180 50 72% <== < 80%
speechbrain/utils/epoch_loop.py 55 22 60% <== < 80%
speechbrain/utils/hparams.py 2 1 50% <== < 80%
speechbrain/utils/hpopt.py 134 41 69% <== < 80%
speechbrain/utils/logger.py 73 45 38% <== < 80%
speechbrain/utils/metric_stats.py 285 48 83%
speechbrain/utils/parameter_transfer.py 87 17 80%
speechbrain/utils/profiling.py 191 54 72% <== < 80%
speechbrain/utils/superpowers.py 20 6 70% <== < 80%
speechbrain/utils/text_to_sequence.py 77 22 71% <== < 80%
speechbrain/utils/torch_audio_backend.py 9 2 78% <== < 80%
speechbrain/utils/train_logger.py 150 113 25% <== < 80%
speechbrain/wordemb/transformer.py 90 67 26% <== < 80%
-----------------------------------------------------------------------------
TOTAL 16782 4481 73%