GitLab

* Separate sox list format function for read and write

* Guard MP3 smoke test

buck gets confused with utility module name with `test_` prefix.

1. Move misplaced sox compatibility test (T,Fade, T.Vol, T.Vad) to test/test_sox_compatibility.py
2. Move test_sox_effects to test/sox_effect/ where all the other functionalities from torchaudio.sox_effects are tested

Currently all the tests in `sox_io_backend` and `sox_effects` (for new SoX effects implementation) requires additional `sox`, and this prevents running test in environment where `sox` command is not available even though `torchaudio` extension is available (such as fb internal). This PR adds smoke tests for these modules, which just runs functions to see if they do not crash.

Co-authored-by: Jeff Zhang <jeffzhang@fb.com>

* Add WaveRNN example

This is the pipeline example based on [WaveRNN model](https://github.com/pytorch/audio/pull/735) in torchaudio. The design of this pipeline is inspired by [#632](https://github.com/pytorch/audio/pull/632). It offers a standardized implementation of WaveRNN vocoder in torchaudio.

* Add utils and readme

The metric logger is added based on the Wav2letter pipeline [#632](https://github.com/pytorch/audio/pull/632). It offers the way to parse the standard output as described in readme.

* Add channel dimension

The channel dimension of waveform in datasets is added to match the input dimensions of WaveRNN model because the channel dimensions of waveform and spectrogram are added in [this part] (https://github.com/pytorch/audio/blob/master/torchaudio/models/_wavernn.py#L281) of WaveRNN model.

* Update date split and transform

The design of dataset structure is discussed in [this comment](https://github.com/pytorch/audio/pull/749#discussion_r454627027

). Now the dataset file has a clearer workflow after using the random-split function instead of walking through all the files. All transform functions are put together inside the transforms block.
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

- Addresses #549 #638 #786 
- Add `torchaudio` top level module doc
- Separate `torchaudio` top level module doc from `index.html`
- Add `backend` module doc.
- Remove `-> None` from function signature as it adds noise to documentation
- Changed function argument name of `torchaudio.backend.sox_io_backend.save` from `tensor` to `src`, so that it matches with the reset of backends.
- Tweak bunch of docstrings

Fix output type of upsampling
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

* Update default form in docstring
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

* Add libritts

Add LibriTTS dataset draft

* Add libritts

Use two separate ids for utterance_id.

* Update output form

Use full_id as utterance_id.

* Update format

Add space and test black format

* Update test method

* Add audio and text test

Generate audio and test files on-the-fly in test 

* Update format

* Fix test error and remove assets libritts

The test error is fixed by sorting the file in 4th element instead of 2nd element in samples. Since the files are generated on-the-fly, so the the libritts files in assets are removed.

* Add seed in `get_whitenoise` function

* Change utterance to text

Change `_utterance` to `_text`.
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

* Change the name of  n_output and n_hidden

* Replace the mode by n_classes and sample_rate

* Change the definition of n_output and n_hidden
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

* Addressed review issues in PR #668

* Changed GTZAN so that it only traverses filenames belonging to the dataset

Now, instead of walking the whole directory and subdirectories of the dataset
GTZAN only looks for files under a `genre`/`genre`.`5 digit number`.wav format, where `genre` is an allowed GTZAN genre label.
This allows moving or removing files from the dataset (e.g. for fixing duplication or mislabeling issues).

* Get rid of sine wave files and whitenoise files
* Refactor integer encoding
* Relax rtol from 1e-8 to 1e-7 for compliance kaldi
* relax waveform multi channel resample atol to 1e-7 from 1e-8
* relax tolerance for length consistency for speed effect
Co-authored-by: moto <855818+mthrok@users.noreply.github.com>

* Add sox_utils module

* Make init/shutdown thread safe

* Add sox effects implementation

* Add test for sox effects

* Update docstrings and add examples

This PR replaces `torchaudio.sox_effects.SoxEffectsChain` in `test_sox_compatibility` with bare `sox` command.

The parity of `torchaudio.sox_effects.SoxEffectsChain` against `sox` command is not tested and it has known issues https://github.com/pytorch/audio/issues/771, therefore it is not appropriate to use this class for testing other functions.

`test/test_sox_effects.py::Test_SoxEffectsChain::test_lowpass_speed` has some issue on our macOS CI, even though there was no issue at #777 .

While we figure out the fix, we disable this test for macOS.

Part of #764

 - Replace `whitenoise.wav` with on-the-fly data generation
 - Replace `torchaudio.load` with `common_utils.load_wav`
 - Replace `steam-train-whistle-daniel_simon.mp3` with `.wav`

In #779, we plan to remove `frames_per_chunk` parameter from `save` function, but it will take some time before we can land #779, so we go ahead and remove the parameter first to reduce the conflict caused by interface change.

* Change 'sox' to 'default'

* converted CommonVoice tartar mp3 to wav using rate 8000 Hz

* Remove Unused dtmf_30s_stereo.mp3

* add Waveforms for Testing Purposes section

* Update test/README.md

use wrapper function for scipy.io.wavfile.read
Co-authored-by: moto <855818+mthrok@users.noreply.github.com>

* remove un-used files from the doc

* Update test/README.md

Rename variable
Co-authored-by: moto <855818+mthrok@users.noreply.github.com>

* fix indent; remove mentions of unused files

* remove whitenoise* files from README.md
Co-authored-by: moto <855818+mthrok@users.noreply.github.com>

* upsamplenetwork

* update variable names

* update variable name

* add wavernn model

* update test

* update format

* update format

* update format

* fix conflicts and add transpose

* import update

* update transpose

* update format

* update docstring

* add n_channel in input

* add comment

* update docstring

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

* Pin llvmlite

* Add comments

* upsamplenetwork

* update name

* update name and docstring

* update format

* rebase

* update docstring

* update docstring

* remove transpose and update docstring
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

This is a part of PRs to add new "sox_io" backend. #726 and depends on #718, #728 and #731.

This PR adds `save` function to "sox_io" backend, which can save Tensor to a file with the following audio formats;
 - `wav`
 - `mp3`
 - `flac`
 - `ogg/vorbis`

* Use cmake for third party

* Apply patch to libmad

* Update gitignore

* Update docker test image

* update varible names and docstring

* update format

* update docsting and output value
Co-authored-by: Ji Chen <jimchen90@devfair0160.h2.fair>

* Update functional.py

rollback torch.norm() in spectrogram() to v0.4.0.

* Update functional.py

comment out `spec_f = complex_norm(spec_f, power=power)`.

* fixed complex_norm() instead of spectrogram() for torch.norm() issue.

* lint
Co-authored-by: bongjin.lee <bongjin.lee@navercorp.com>
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

This is a part of PRs to add new "sox_io" backend. #726 and depends on #718 and #728 .

This PR adds `load` function to "sox_io" backend, which is  tested on the following audio formats;
 - `wav`
 - `mp3`
 - `flac`
 - `ogg/vorbis` *

By default, "sox_io" backend returns Tensor with `float32` dtype and the shape of `[channel, time]`. The samples are normalized to fit in the range of `[-1.0, 1.0]`.

Unlike existing "sox" backend, the new `load` function can handle WAV file natively, when the input format is WAV with integer type, (such as 32-bit signed integer, 16-bit signed integer and 8-bit unsigned integer) by providing `normalize=False`, this function can return integer Tensor, where the samples are expressed within the whole range of the corresponding dtype, that is, `int32` tensor for `32-bit PCM`, `int16` for `16-bit PCM` and `uint8` for `8-bit PCM`. This behavior follows [scipy.io.wavfile.read](https://docs.scipy.org/doc/scipy/reference/generated/scipy.io.wavfile.read.html). `normalize` parameter has no effect for other formats and the load function always return normalized value with `float32` Tensor.

__* Note__ The current binary distribution of torchaudio does not contain `ogg/vorbis` and `opus` codecs. To handle these files, one needs to build torchaudio from the source with proper codecs in the system.

__Note 2__ Since this PR, `scipy` becomes required module for running test.