GitLab

Summary:
# Changes
* Adding workflow for building aarch64 wheels.

Pull Request resolved: https://github.com/pytorch/audio/pull/3553

Reviewed By: hwangjeff, osalpekar

Differential Revision: D48239384

Pulled By: atalman

fbshipit-source-id: dfa00edb3fee0acaf2b83fb420eaf12bddc6980e

Summary:
The backend dispatcher is implemented in `torchaudio._backend`, while the legacy backend is implemented in `torchaudio.backend`.

The initialization happen in `torchaudio._backend`.
This commit moves it to `torchaudio.__init__`, so that `backend` and `_backend` is more independent.

Pull Request resolved: https://github.com/pytorch/audio/pull/3548

Reviewed By: huangruizhe

Differential Revision: D48219244

Pulled By: mthrok

fbshipit-source-id: e694cb232794f90902a60ee51c7bf11b7f0548a0

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3550

Reviewed By: hwangjeff

Differential Revision: D48219176

Pulled By: mthrok

fbshipit-source-id: 4b11111dd3853cbef4ffe1859ec428ca05394824

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3546

Reviewed By: huangruizhe

Differential Revision: D48219274

Pulled By: mthrok

fbshipit-source-id: 6881f039bf70cf7240fbcfeb48443471ef457bd4

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3544

Revises VGGish inference pipeline test to support internal testing.

Reviewed By: mthrok

Differential Revision: D48058409

fbshipit-source-id: 045140a0e9d50128d32ef6510bdb2f642a365c83

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3542

Reviewed By: huangruizhe

Differential Revision: D48166025

Pulled By: mthrok

fbshipit-source-id: 29fee7dbf08394993972ec2967f94ce9fcb1c853

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3532

Reviewed By: mthrok

Differential Revision: D48165499

Pulled By: mpc001

fbshipit-source-id: c87b3361f0e6282684f218b32888df883d56682b

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3534

Reviewed By: huangruizhe

Differential Revision: D48155817

Pulled By: mthrok

fbshipit-source-id: a3d45fdfd360f9668063a3ecb3b00364290134c9

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3538

Reviewed By: huangruizhe

Differential Revision: D48154056

Pulled By: mthrok

fbshipit-source-id: 72f58c501c5302d40f1d059f95bd6fe40d4a52aa

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3336

Reviewed By: mthrok

Differential Revision: D47846814

Pulled By: huangruizhe

fbshipit-source-id: dc12362bf243c52222dccadec3176e25e43dd652

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/1959

Reviewed By: hwangjeff

Differential Revision: D32078361

fbshipit-source-id: 50b56bac9593c36197998e89db19cd6d65b793cc

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3536

Reviewed By: huangruizhe

Differential Revision: D48120170

Pulled By: mthrok

fbshipit-source-id: dec7575db07734490099b35a8bfc854252952c6e

Summary:
Port the MMS FA model from tutorial to the library with post-processing module.

Pull Request resolved: https://github.com/pytorch/audio/pull/3521

Reviewed By: huangruizhe

Differential Revision: D48038285

Pulled By: mthrok

fbshipit-source-id: 571cf0fceaaab4790983be2719f1a85805b814f5

Summary:
This commit adds `merge_tokens` function which removes repeated tokens from CTC token sequences returned from `forced_align`.

Resolving repeated tokens is a necessary step and almost universal, thus it makes sense to have such helper function in torchaudio.

Pull Request resolved: https://github.com/pytorch/audio/pull/3535

Reviewed By: huangruizhe

Differential Revision: D48111202

Pulled By: mthrok

fbshipit-source-id: 25354bfa210aa5c03f8c1d3e201f253ca3761b24

Summary:
Currently `torchaudio.functional.forced_align` function requires full information on input/target lengths.
When performing non-batched alignment, these can be inferred from the size of Tensor.

Pull Request resolved: https://github.com/pytorch/audio/pull/3533

Reviewed By: nateanl

Differential Revision: D48111041

Pulled By: mthrok

fbshipit-source-id: fbf07124d3959c5cc5533dcd86296851587082fb

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/3531

Revises VGGish pipeline to accept arbitrary state dict function to accommodate loading weights from any source.

Reviewed By: mthrok

Differential Revision: D48056390

fbshipit-source-id: 2767699b58442ad132b518b4a6435f2772a637c3

Summary:
- Simplify the step to generate token-level alignment

Pull Request resolved: https://github.com/pytorch/audio/pull/3529

Reviewed By: huangruizhe

Differential Revision: D48066787

Pulled By: mthrok

fbshipit-source-id: 452c243d278e508926a59894928e280fea76dcc6

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3527

Reviewed By: huangruizhe

Differential Revision: D48008822

Pulled By: mthrok

fbshipit-source-id: 4beae2956dfd1f00534832b70a1bf0897cba7812

Summary:
Increases numerical tolerance on Conformer RNN-T TorchScript consistency tests to resolve CI test failures.

Pull Request resolved: https://github.com/pytorch/audio/pull/3525

Reviewed By: mthrok

Differential Revision: D48000613

Pulled By: hwangjeff

fbshipit-source-id: 1d35ba58055a8346dc40e2b67f37ccfd2e015894

Summary:
When passing int16 type tensor to `save(backend="sox")`, the resulting file should be 16-bit signed PCM, but instead is 32-bit signed PCM.

Resolves https://github.com/pytorch/audio/issues/3304

Pull Request resolved: https://github.com/pytorch/audio/pull/3524

Reviewed By: huangruizhe

Differential Revision: D47941090

Pulled By: mthrok

fbshipit-source-id: 2622b31eb1cbf03969f67ab2b2adec6e2ba677c4

Summary:
Add a separate tutorial for cuctc.
Reslove https://github.com/pytorch/audio/issues/3096

Pull Request resolved: https://github.com/pytorch/audio/pull/3297

Reviewed By: huangruizhe

Differential Revision: D47928400

Pulled By: mthrok

fbshipit-source-id: 8c16492fb4d007b6ea7969ba77c866a51749c0ec

Summary:
torch.nn.utils.weight_norm is deprecated.
Replacing this with new API

Pull Request resolved: https://github.com/pytorch/audio/pull/3523

Reviewed By: huangruizhe

Differential Revision: D47932384

Pulled By: mthrok

fbshipit-source-id: 344abfa12bd11da779f7fd13b74a1e009a582b52

Summary:
Adds pre-trained VGGish inference pipeline ported from https://github.com/harritaylor/torchvggish and https://github.com/tensorflow/models/tree/master/research/audioset.

Pull Request resolved: https://github.com/pytorch/audio/pull/3491

Reviewed By: mthrok

Differential Revision: D47738130

Pulled By: hwangjeff

fbshipit-source-id: 859c1ff1ec1b09dae4e26586169544571657cc67

Summary:
torch.norm is now deprecated.
The usages in torchaudio seems to be vector norm, so replacing them with torch.linalg.vector_norm

Resolves https://github.com/pytorch/audio/issues/3484

Pull Request resolved: https://github.com/pytorch/audio/pull/3522

Reviewed By: huangruizhe

Differential Revision: D47926659

Pulled By: mthrok

fbshipit-source-id: f7428cf0168109a3d340b8784adc99bb5f781084

Summary:
- Set global matplotlib rc params
- Fix style check
- Fix and updates FA tutorial plots
- Add av-asr index cars

Pull Request resolved: https://github.com/pytorch/audio/pull/3515

Reviewed By: huangruizhe

Differential Revision: D47894156

Pulled By: mthrok

fbshipit-source-id: b40d8d31f12ffc2b337e35e632afc216e9d59a6e

Summary:
The I/O functions in _compat module was introduced there so that
everything related to FFmpeg is in torchaudio.io and FFmpeg library
initialization can be carried out in `torchaudio.io.__init__`.

Now that this constraint is removed, (all the initialization happens
at `torchaudio._extension.__init__`) and `_compat` is only used by
FFmpeg dispatcher backend, we move the module to `torchaudio._backend`
for better locality.

Pull Request resolved: https://github.com/pytorch/audio/pull/3518

Reviewed By: huangruizhe

Differential Revision: D47877412

Pulled By: mthrok

fbshipit-source-id: aa18c8cb6e5d5360950df5158c33c653e37c565f

Summary:
Context: https://github.com/pytorch/audio/issues/3448

The documentation of amplitude_to_DB is ambigious on how cut-off values are computed when the input tensor is 3D.

This commit clarifies that.

Closes: https://github.com/pytorch/audio/issues/3448

Pull Request resolved: https://github.com/pytorch/audio/pull/3519

Reviewed By: huangruizhe

Differential Revision: D47875505

Pulled By: mthrok

fbshipit-source-id: e06bb997e7a27e2abe35c8e2ac91ddfbded4e641

Summary:
In https://github.com/pytorch/audio/issues/2419, we added ffmpeg as fallback for sox_io backend. The was a warkaround for solving the issue with libmad removal.

Now that we introduced `backend` argument to I/O functions, and libsox integration is moved to dynamic binding where users can use libsox with libmad integration, we do not need the workaround.

This commit is based on reverting https://github.com/pytorch/audio/issues/2416 (fd7ace17).

Pull Request resolved: https://github.com/pytorch/audio/pull/3516

Reviewed By: huangruizhe

Differential Revision: D47855272

Pulled By: mthrok

fbshipit-source-id: 5af73af7865f6e545ccb052d478e86588ff2a014

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3517

Reviewed By: huangruizhe

Differential Revision: D47858452

Pulled By: mthrok

fbshipit-source-id: 62ee6c8bb2669dd70f8ca25703a04dc8a9d19aec

Summary:
The PR move `SquimObjective` and `SquimSubjective` models and corresponding factory functions and pre-trained pipelines out of prototype and to the core directory. They will be included in the next official release.

Pull Request resolved: https://github.com/pytorch/audio/pull/3512

Reviewed By: mthrok

Differential Revision: D47837434

Pulled By: nateanl

fbshipit-source-id: d0639f29079f7e1afc30f236849e530c8cadffd8

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3511

Reviewed By: mthrok

Differential Revision: D47852108

Pulled By: mpc001

fbshipit-source-id: c0ecb4b5bcc8670013dcbe1164e3929f5793c8aa

Summary:
Removes residual from https://github.com/pytorch/audio/issues/3497

Pull Request resolved: https://github.com/pytorch/audio/pull/3514

Differential Revision: D47838049

Pulled By: mthrok

fbshipit-source-id: c4b00aba9f4cc887ec595f04d7a2dd673c63b975

Summary:
This commit updates the way libsox is integrated to torchaudio

1. We stop statically linking libsox, so torchaudio will not ship libsox.
2. We link libsox dynamically. Users are expected to install libsox by themselves.
3. We use stab library to build torchaudio.

Pull Request resolved: https://github.com/pytorch/audio/pull/3497

Differential Revision: D47803706

Pulled By: mthrok

fbshipit-source-id: 31b05495d81069186fa52d67beea360cc7e817a8

Summary:
Since libsox and ffmpeg extensions now depend on external libraries, their initialization processes might cause unrecoverable issue, such as segfault.

This commit adds environment variable to disable them so that importing torchaudio won't attempt to load these libraries.

Pull Request resolved: https://github.com/pytorch/audio/pull/3500

Differential Revision: D47808178

Pulled By: mthrok

fbshipit-source-id: 80c1c6b5f4bc608d4e209473702680db093c95ee

Summary:
This PR moves video loading outside detector during pre-processing.

Pull Request resolved: https://github.com/pytorch/audio/pull/3498

Reviewed By: mthrok

Differential Revision: D47811044

Pulled By: mpc001

fbshipit-source-id: f17839b695b13d3cf2d9db343d7e9a0202eea7d5

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3499

Differential Revision: D47803654

Pulled By: mthrok

fbshipit-source-id: 2b916fa66d84c91c01b4dfe6dd5ee3501159f451

Summary:
Add scheduled doc update job so that docs are updated at least once a day.

Pull Request resolved: https://github.com/pytorch/audio/pull/3496

Differential Revision: D47795577

Pulled By: mthrok

fbshipit-source-id: aba5376ec51f07560014d250a16fef8b8a11b43e

Summary:
In preparation for https://github.com/pytorch/audio/pull/3082

Disable those FFmpeg tests that depend on sox CLI. These tests need to be updated or removed so as not to use sox CLI.

Auto-skip some sox tests if decoder/encoder are not available

Pull Request resolved: https://github.com/pytorch/audio/pull/3494

Differential Revision: D47761948

Pulled By: mthrok

fbshipit-source-id: 3a48d7f280f8376a48d223947dd41a7cdc8cbc30

Summary:
- Fix condition to add new commit to gh-pages
- Allow to deploy docs from workflow dispatch

Pull Request resolved: https://github.com/pytorch/audio/pull/3495

Differential Revision: D47767443

Pulled By: mthrok

fbshipit-source-id: 9ca858868f3e822e532c21cde9d7499af9891a51

Summary:
This PR is to include few changes in the AV-ASR recipe. The changes include better results, a faster face detector (Mediapipe), renamed variable names, a streamlined dataloader, and a few illustrated examples. These changes were made to improve the usability of the recipe.

Pull Request resolved: https://github.com/pytorch/audio/pull/3493

Reviewed By: mthrok

Differential Revision: D47758072

Pulled By: mpc001

fbshipit-source-id: 4533587776f3a7a74f3f11b0ece773a0934bacdc