GitLab

* Suggested changes to speed up phaser

* Checkpoint

* Checkpoint

* Checkpoint

* Checkpoint

* removing todo items
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

As a part of the "sox" backend sunset plan (#903), we add a "soundfile" backend that is compatible with the "sox_io" backend. No new public backend name is added. We provide a switch to change the interface/behavior of "soundfile" backend.

This commit contains;
 - The implementation of the new "soundfile" backend.
 - The flag to switch the behavior of "soundfile" backend. (`torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE`)
 - Test for the new backend and switching mechanism.

The default behavior of "soundfile" backend is not changed. The users who want to opt-in the new "soundfile" interface can do so by `torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = False` before changing the backend to "soundfile".

In 0.8.0 release, the "soundfile" backend will use this interface by default, and users can still use the legacy one with `torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = True`. In 0.9.0, the legacy interface is removed and `torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE` flag will be eventually removed.

* example pipeline, initial commit.

* removing notebook conversion artifacts.

* remove extra comments. lint.

* addressing some feedback.

* main function.

* defining args in function.

* refactor.

* lint.

* checkpoint.

* clean version to start with.

* adding more parameters.

* lint.

* cleaning full version.

* check for not None.

* cleaning.

* back -l 160

* black.

* fix runtime error.

* removing some print statements.

* add help to command line. add progress bar option.

* grouping librispeech-specific transform in subclass.

* typo.

* fix concatenation.

* typo.

* black. tqdm.

* missing transpose.

* renaming variables.

* sum cer and wer

* clip norm.

* second signal handler removed.

* cosmetic.

* default to no checkpoint.

* remove non_blocking.

* adadelta works better than sgd.

* anomaly detection.

* moving dataset to separate file.

* lint.

* move to separate module: languagemodel, decoder, metric.

* flush=True.

* renaming decoder.

* CTC Decoders.

* flush=True.

* pass length for viterbi decoder.

* progress bar. relative path.

* generalize transition matrix to n-gram. progress bar.

* choice of decoder.

* collate func.

* remove signal handling.

* adding distributed.

* lint.

* normalize w/r to length of dataset, and w/r to total number characters.

* relative cer/wer.

* clip grad parameter. momentum back but not yet used.

* Switch to SGD.

* choice of optimizer.

* scheduler.

* move to utils file.

* metric log, and utils file.

* rename metric_logger.

* stderr and stdout. simpler metric logger.

* replace by logging.

* adding time measurement in metric logger.

* fix duplicate name. remove tqdm. keep track of epoch instead and iteration instead.

* rename main file. and add readme.

* refactor distributed.

* swap example and output in readme.

* remove time from logger.

* check non-empty tensor input.

* typo in variable name and log update.

* typo.

* compute cer/wer in training too.

* typo.

* add back slurm signal capture to resubmit job.

* update levinstein distance.

* adding tests for levenstein distance.

* record error rate during iteration.

* metric logger using setitem.

* moving signal break to end of loop and return loss so far.

* typo.

* add citation.

* change default to best run.

* adding other experiment with decoders.

* remove other decoders than greedy.

* Revert "remove other decoders than greedy."

This reverts commit fb114372e89e317bf48d0b1f846c60bca8efe1ac.

* changing name of folfder.

* remove other decoders, and unused dataset class.

* rename functions to align with other pipeline.

* pick which parts to train with.

* adding specaugment to validation. note that caching prevents randomization from happening in validation.

* updating readme.

* typo in metric logging.

* Revert "typo in metric logging."

This reverts commit acac245eec250f61d2039a67933d3c01f1975ce9.

* Revert "Revert "typo in metric logging.""

This reverts commit 2c80d9691ed401044da734c40df3715dba92d0db.

* update metric logger.

* simplify metric logger implementation.

* use json dumps instead.

* group metric together.

* move function.

* lint.

* quick summary of files in folder.

* pass clip_grad explictly.

* typo in default dataset name.

* option to disable logger.

* ergonomics for distributed.

* reminder about signal handler.

* minor refactor of main in main.

* replace by not_main_rank.

* raising error if parameter not supported.

* move model before invoking DDP.

* changing log level. using python 2 style string for logging.

* dynamic augmentations.

* update metric log.

batch cer/wer metric. correct typo in time. adding other dimensions in metric.

* save learning rate even if function not available.

* add type option to model.

* add adamw.

* reduce lr on validation step or training step.

* specify hop-length and win-length.

* normalize option.

* rename parameter.

* add dropout and tweak to number of channels.

* copy model in pipeline folder for experimentation.

* fix scheduler stepping.

* fix input_type and num_features.

* waveform mode changes shape more.

* adding best character error rate with current implementation of model with mfcc.

* comment update.

* remove signal. remove custom wav2letter model.

* remove comment.

* simpler import with pandas.

This reverts commit 1ecbc249.

Reason: The macOS CI jobs are stuck with xdist recently.

* Warn if create_fb_matrix produces a column whose weights are all zeros

See also https://github.com/librosa/librosa/issues/478

Default to using the `cpu` directory since the base one is not reliable
Signed-off-by: Eli Uriegas <eliuriegas@fb.com>

* Add deprecation warning to sox backend

Refer to https://github.com/pytorch/audio/issues/903

* updated the build_generator call to include the models argument

* fixed RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same

Upstream is on 9.4.1 and we were experiencing issues when conda-build
was attempting to check for overlinking and failed out due to some files
not existing in xcode 9.0
Signed-off-by: Eli Uriegas <eliuriegas@fb.com>

When running smoke test, proper way to install conda package is to
install from workspace, instead of installing from archive.
Otherwise the dependencies are not properly installed

https://github.com/conda/conda/issues/1884

For this, we pass the whole `conda-bld` directory from
conda package build job to upload job then smoke test job.

`torchaudio` does not have a mandatory third-party python package dependency,
so this was not an issue.

See also https://github.com/pytorch/text/pull/803

* Fix incorrect extension parsing in sox_io_backend.save
* Add tests for compression=None

* Tweak docstring, audio_ext, load method signature and constructor of VCTK_092

* Add test for VCTK_092 dataset.

* Added version 0.92 of VCTK dataset
Signed-off-by: Abhishek Dubey <abhi.dubey011999@gmail.com>

librosa 0.8.0 has been released which include fix for
https://github.com/librosa/librosa/issues/1160 at
https://github.com/librosa/librosa/pull/1164

* Add spectrogram normalization option
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

This CC adds CI test to check the backward compatibility of Torchscript functions/object dumps.

The job first dumps Torchscript objects into file from torchaudio 0.6.0 release environment (x Python 3.6, 3.7, 3,8), then load & run the function in master build (in Python 3.6, 3.7, 3.8).

If there is a BC-breaking change in master build, (registration schema change), then the test should fail.

At this moment, `info` function does not work due to suspected bug in torch side, so the test is disabled for `info` function. See https://github.com/pytorch/pytorch/issues/42258 for the detail. Once https://github.com/pytorch/pytorch/issues/42258 is resolved we can enable it.

We have been running unit test with editable installation. (i.e. `python setup.py develop`), with which we missed issues like #842. 

This CC makes installation in CI non-editable, and change test directory structure so that the source code will not shadow the installed version of `torchaudio`. With simple `pytest test`, `pytest` modifies `sys.path` and prepend checked out repository, which shadows the installed version.

To remedy this, the whole test suites has been moved from `./test` to `./test/torchaudio_unittest`. This adds nice module structure to our test code and we can do absolute import in each test module, which makes it possible again to run test with `python -m unittest torchaudio_unittest/XXX.py`

This change does not affect the regular development process (`python setup.py develop` && `pytest test`)

Currently `setuptools` assumes that torhcaudio is zip safe and performs
egg (zip) installation when `python setup.py install` (or `pip install .`).

When `torchaudio` loads extension module written with Torchscript, 
the corresponding loading function requires the `so` file to be present as 
an actual file, and this does not work well with egg installation.

This CC fixes this by setting `zip_safe=False` in setup.py and disable egg
installation, so that `python setup.py install` installs uncompressed files
with regular directory structure.

Currently our test suites automatically/silently skip tests on C++ extension
if it is not available. This is nice in local env, but in CI these tests should be
enforced and reported as failure if C++ extension is not available.

This PR adds switch for making tests fail if C++ extension is not available,
and make CI for *nix fail if that's the case.

setuptools warns that it should be imported before Distutils

```
site-packages/setuptools/distutils_patch.py:25: UserWarning: Distutils was imported before Setuptools. This usage is discouraged and may exhibit undesirable behaviors or errors. Please use Setuptools' objects directly or at least import Setuptools first.
```

* Update links
* Add how to run tests
* Update asset info

* `istft` has been migrated to `pytorch`, and `torchaudio.functional.istft` has been deprecated in 0.6.0 release. This PR removes it
Co-authored-by: Jeremy Chen <jeremyyy@fb.com>

Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

Currently macOS CI job takes twice the time of linux unit tests.
This PR uses `pytest-xdist` to run unit tests in parallel on macOS CI job. 

Interestingly applying the same technique to linux unit tests make the test execution longer.
So the change is only applied to macOS.

macOS 27 mins -> 17 mins 😸
linux 14 mins -> 1 hours 😾

Now that most of our test cases are backend aware/agnostic,
the use of BACKENDS and BACKENDS_MP3 is anti-pattern.

This change moves these definitions to the legacy test module `test_io`.

If a new backend is added in future, that backend should have separate test suite,
(like the ones in `test/sox_io_backend`) so instead of dynamically listing available
backend in `test_io`, the changed code only checks "sox" backend and "soundfile" backend.