GitLab

Summary:
This reverts commit b7d3e89a.

We will use pre-built binaries instead of dlopen.

Pull Request resolved: https://github.com/pytorch/audio/pull/3456

Differential Revision: D47239681

Pulled By: mthrok

fbshipit-source-id: 0446a62410d914081184fc20c386afa00b1e41b6

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3455

Differential Revision: D47242316

Pulled By: mthrok

fbshipit-source-id: 0eb4bdb0a45fccfe9ff97eaed79db63cd7bfc7d8

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3457

Differential Revision: D47241343

Pulled By: mthrok

fbshipit-source-id: fd1bfd1531397cb59e9cf11de9dede6949f8517e

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3433

Current design of forced_align accept 2D Tensor for `log_probs` and 1D Tensor for `targets`. To make the API simple, the PR make changes to only support batch Tensors (3D Tensor for `log_probs` and 2D Tensor for `targets`).

Reviewed By: mthrok

Differential Revision: D46657526

fbshipit-source-id: af17ec3f92f1a2c46dba91c6db2488a11de36f89

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3434

Add one bullet point for `torchaudio.functional` and forced alignment as one example.

Reviewed By: mthrok

Differential Revision: D46658058

fbshipit-source-id: 6e037b7bb6ed2fc2e27ad1e55c5728c17ce69ce8

Summary:
Include Conformer/Emformer RNN-T ASR/VSR/AV-ASR link to index.rst

Pull Request resolved: https://github.com/pytorch/audio/pull/3441

Differential Revision: D47094158

Pulled By: mthrok

fbshipit-source-id: 9ab42ac2bf52a5ce488003897ffba2f10a6ca941

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3449

Differential Revision: D47094402

Pulled By: mthrok

fbshipit-source-id: 43e6994604f0e6c06a5f19c5e8599e2ce12ae622

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3442

Differential Revision: D46797481

Pulled By: mthrok

fbshipit-source-id: 3513037cbb8f2edb70fdab0fec5c7c554a697abe

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3427

Adds transform `ChromaSpectrogram` for generating chromagrams from waveforms as well as transform `ChromaScale` for generating chromagrams from linear-frequency spectrograms.

Reviewed By: mthrok

Differential Revision: D46547418

fbshipit-source-id: 250f298b8e11d8cf82f05536c29d51cf8d77a960

Summary:
Splitting the multilingual example part into another tutorial.

Pull Request resolved: https://github.com/pytorch/audio/pull/3443

Reviewed By: mthrok

Differential Revision: D46802844

Pulled By: xiaohui-zhang

fbshipit-source-id: a7093053cac8b79d650d4f665db7fde2d8254998

Summary:
This PR adds a data preparation recipe that uses the ultra face detector to extract full-face video. The resulting video output is then used as input for training and evaluating RNNT-based models for automatic speech recognition (ASR), visual speech recognition (VSR), and audio-visual ASR (AV-ASR) on the LRS3 dataset.

This PR also updates the word error rate (WER) for AV-ASR LRS3 models and improves the code readability.

Pull Request resolved: https://github.com/pytorch/audio/pull/3421

Reviewed By: mpc001

Differential Revision: D46799748

Pulled By: mthrok

fbshipit-source-id: 97af3feac0592b240617faaffa4c0ac8cef614a9

Summary:
* Fix backtrack visualization (the cooridnate was off-by-one.)
* Add note about the simplification and the new align API
* Explicitly handle SOS and EOS

Pull Request resolved: https://github.com/pytorch/audio/pull/3440

Reviewed By: xiaohui-zhang

Differential Revision: D46761282

Pulled By: mthrok

fbshipit-source-id: b0b6c9754674e8e23543e9f002e29b55102c92f8

Summary:
Currently, AudioEffector always resample to the original sample rate. It is more flexible to allow overriding this to any sample rate.

Pull Request resolved: https://github.com/pytorch/audio/pull/3374

Differential Revision: D46235358

Pulled By: mthrok

fbshipit-source-id: 39a5d4e38d9b90380da31d0ce9ee8090668b54e4

Summary:
This plugin pulls glib and gtk which breaks the build on some headless systems

Since the plugin is not actually used, it seems right to disable it

This change fixed the build on my system

Pull Request resolved: https://github.com/pytorch/audio/pull/3436

Differential Revision: D46683297

Pulled By: mthrok

fbshipit-source-id: 5b1c1eee1929f4a69a1cc6c7d7bb3ed998ec5872

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3435

Reviewed By: nateanl

Differential Revision: D46659362

Pulled By: mthrok

fbshipit-source-id: ffa033ad6759de6fd958b63ac51a4a1153ffb45d

Summary:
Should resolve https://github.com/pytorch/audio/issues/3425

cc mthrok

Pull Request resolved: https://github.com/pytorch/audio/pull/3432

Differential Revision: D46656180

Pulled By: mthrok

fbshipit-source-id: 5c534bee2f143ef5cb5e50ec74828012dbcab7e9

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3422

Differential Revision: D46558184

Pulled By: mthrok

fbshipit-source-id: a775c4fb193496d9b2bf9db7bee186ee23512b99

Summary:
The new version of transformers changed the format of pre-trained weight. Fixing it is low-priority for the maintanance team so we disable the test.

See https://github.com/pytorch/audio/issues/3430

Pull Request resolved: https://github.com/pytorch/audio/pull/3431

Differential Revision: D46592883

Pulled By: mthrok

fbshipit-source-id: d8f54a281a92cac60c469c48f95345bcf0e959d6

Summary:
StreamWriter's encoding pipeline looks like the following

1. convert tensor to AVFrame
2. pass AVFrame to AVFilter
3. pass the resulting AVFrame to AVCodecContext (encoder) and AVFormatContext (muxer)

When dealing with CUDA tensor, the AVFilter becomes no-op, as we have not added support for CUDA-compatible filters.

When CUDA frame is passed, the existing solution passes the software pixel format to AVFilter, which issues warning later as what AVFilter sees is AV_PIX_FMT_CUDA.

Since the filter itself is no-op, it functions as expected. But this commit fixes it.

See https://github.com/pytorch/audio/issues/3317

Pull Request resolved: https://github.com/pytorch/audio/pull/3426

Differential Revision: D46562370

Pulled By: mthrok

fbshipit-source-id: ce0131f1e50bcc826ee036fc0f35db2a5162b660

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3395

Adds chroma filter bank function `chroma_filterbank` to `torchaudio.prototype.functional`.

Reviewed By: mthrok

Differential Revision: D46307672

fbshipit-source-id: c5d8104a8bb03da70d0629b5cc224e0d897148d5

Summary:
[Nova] Add cache ffmpeg before building - 2
Follow up after https://github.com/pytorch/audio/pull/3417, need to pass new arguments to test-infra workflows

Pull Request resolved: https://github.com/pytorch/audio/pull/3423

Reviewed By: mthrok

Differential Revision: D46559344

Pulled By: atalman

fbshipit-source-id: fa5cccc3bfb052688de4a05cc3b4f37fcbe3a6f5

Summary:
StreamReader decoding process is composed of the three steps;

1. Decode the incoming AVPacket into AVFrame
2. Pass AVFrame through AVFilter to perform post process
3. Convert the resulgint AVFrame

The internal of StreamReader was refactored in https://github.com/pytorch/audio/issues/3188 so that the above pipeline is initialized at the time output stream is defined and output stream shape can be retrieved.

For CPU decoder, this works fine because resizing happens in step 2, and the resulting shape can be retrievable.
However, this is problematic for GPU decoder, as resizing is currently done using GPU decoder option (step 1) and there seems to be no interface to retrieve the output shape. This refactor introduced regression, which is described in https://github.com/pytorch/audio/issues/3405

AVFilter internally is adoptive to the change of input frame size. This commit changes the conversion process to be similar, so that it will wait until the first frame comes in to finalize the frame shape.

Fix https://github.com/pytorch/audio/issues/3405

Pull Request resolved: https://github.com/pytorch/audio/pull/3419

Differential Revision: D46557505

Pulled By: mthrok

fbshipit-source-id: 46ad2d82c8c30f368ebfbaf6947718a5036c7dc6

Summary:
CI jobs are migrated from CCI to GHA

Pull Request resolved: https://github.com/pytorch/audio/pull/3420

Differential Revision: D46548562

Pulled By: mthrok

fbshipit-source-id: d7e17201e8b256efaa54543e445a0f139aa549b2

Summary:
- Moving the unit test scripts from .circleci to .github
- Remove docker file for unit test base
- Use the Conda from Docker image in Linux jobs.

Remaining follow-up items

- Reuse the unittest script in Linux GPU job like done in Linux CPU job.

The unit test script needs to be fixed to be used for Linux GPU job
in new GHA workflow. Keeping it as a separate follow-up work item.

Pull Request resolved: https://github.com/pytorch/audio/pull/3407

Differential Revision: D46498263

Pulled By: mthrok

fbshipit-source-id: d8256717a55bb4257151d819d3b2ebd453601eac

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3382

The voice activity detector function was unoptimized, confusingly written, and buggy.

The optimizations created here allow for the function to run roughly 17x faster.
The main optimizations were to loop over windows of audio rather than individual audio samples. Reducing the number of copies also helped.

There was an off by one error where the array slice referenced was [1: 16001] (for the default settings) instead of [0: 16000]

Reviewed By: hwangjeff

Differential Revision: D44749359

fbshipit-source-id: c76c9412e70cdc6fcd527d113603c88f78480558

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3414

Differential Revision: D46536717

Pulled By: mthrok

fbshipit-source-id: 505bdcdd1b59ca9fe5afc2c8516a0a821e2b8d7e

Summary:
[Nova] Add cache ffmpeg before building

Pull Request resolved: https://github.com/pytorch/audio/pull/3417

Reviewed By: mthrok

Differential Revision: D46537892

Pulled By: atalman

fbshipit-source-id: 9f8dc0ecfc305c3b378557d46f89a5d7de67a165

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3415

Differential Revision: D46526437

Pulled By: mthrok

fbshipit-source-id: f78d19c19d7e68f67712412de35d9ed50f47263b

Summary:
To investigate https://github.com/pytorch/audio/issues/3411

Pull Request resolved: https://github.com/pytorch/audio/pull/3418

Differential Revision: D46535891

Pulled By: mthrok

fbshipit-source-id: b90bba399eb54f9f0ae073bd590cd8a46054ed7e

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3410

Differential Revision: D46496786

Pulled By: mthrok

fbshipit-source-id: e517b273c40b340f39ce7db7ab1be1c3eb5f2059

Summary:
We are using Project Nova workflows now. These are not required.

Same as: https://github.com/pytorch/vision/pull/7656

Pull Request resolved: https://github.com/pytorch/audio/pull/3409

Reviewed By: mthrok

Differential Revision: D46494331

Pulled By: atalman

fbshipit-source-id: a642ae55b75482918e0afb7c55dc876bc8356e70

Differential Revision:
D46126226

Original commit changeset: 42cb52b19d91

Original Phabricator Diff: D46126226

fbshipit-source-id: 372b2526d9e196e37e014f1556bf117d29bb1ac6

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3365

Current design of forced_align accept 2D Tensor for `log_probs` and 1D Tensor for `targets`. To make the API simple, the PR make changes to only support batch Tensors (3D Tensor for `log_probs` and 2D Tensor for `targets`).

Reviewed By: vineelpratap

Differential Revision: D46126226

fbshipit-source-id: 42cb52b19d91bbff7dc040ccf60350545d75b3a2

Summary:
Follow up of: https://github.com/pytorch/audio/pull/3368

Remove files and lines no longer used.

Pull Request resolved: https://github.com/pytorch/audio/pull/3403

Differential Revision: D46441462

Pulled By: mthrok

fbshipit-source-id: 11b881ec4b24fa0d625c6aee9f4bd91f637f9923

Summary:
There are some BC-Breaking changes from pytorch_lightning to lightning library. The PR adjust those changes to support latest lightning library.

Pull Request resolved: https://github.com/pytorch/audio/pull/3396

Reviewed By: mthrok

Differential Revision: D46345206

Pulled By: nateanl

fbshipit-source-id: 59469c15dc5fe5466a99a5b5380eb4f98c2c633f

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3402

This is a second attempt of https://github.com/pytorch/audio/pull/3353.

The basic logic to enable dlopen for FFmpeg libraries are same.
It uses `at::DynamicLibrary`, which allows to compile torchaudio without
linking FFmpeg libraries.

This time, the option to enable this feature DLOPEN_FFMPEG has been added,
so that users have a way to disable this feature and keep using build-time
linking.

Please refer to stub.h for more technical detail.

Differential Revision: D46403783

fbshipit-source-id: ca3db57ff6bdc50c8c225d22f12f3e76c6dc3f16

Summary:
This commit removes compute_kaldi_pitch function and the underlying Kaldi integration from torchaudio.

Kaldi pitch function was added in a short period of time by integrating the original Kaldi implementation, instead of reimplementing it in PyTorch.

The Kaldi integration employed a hack which replaces the base vector/matrix implementation of Kaldi with PyTorch Tensor so that there is only one blas library within torchaudio.

Recently, we are making torchaudio more lean, and we don't see a wide adoption of kaldi_pitch feature, so we decided to remove them.

See some of the discussion https://github.com/pytorch/audio/issues/1269

Pull Request resolved: https://github.com/pytorch/audio/pull/3368

Differential Revision: D46406176

Pulled By: mthrok

fbshipit-source-id: ee5e24d825188f379979ddccd680c7323b119b1e

Summary:
Replace sox_effects with `torchaudio.io.AudioEffector`

1. To show case the new and better feature
2. To prepare for the upcoming removal of file-like support object

Pull Request resolved: https://github.com/pytorch/audio/pull/3375

Reviewed By: nateanl

Differential Revision: D46379016

Pulled By: mthrok

fbshipit-source-id: 70f24b62494204949f327f6ac6c49f315c9ee315

Differential Revision:
D46059199

Original commit changeset: 4493a5fd8a4c

Original Phabricator Diff: D46059199

fbshipit-source-id: 71cde3f8cd870d1ad9114e3e87cdd1ba564441c0

Summary:
This commit changes the way FFmpeg extension is built and used.
Instead of linking (LGPL) FFmpeg libraries to torchaudio at build time,
It uses dlopen to search and link them at run time.

For dlopen-ing, we use PyTorch's `at::DynamicLibrary` class, which provides
portable wrapper.

Pull Request resolved: https://github.com/pytorch/audio/pull/3353

Differential Revision: D46059199

Pulled By: mthrok

fbshipit-source-id: 4493a5fd8a4c802178d20276522f5334d637307d