GitLab

Summary:
Part 1 of [T138011314](https://www.internalfb.com/intern/tasks/?t=138011314)

This PR ports the generator part of [HiFi GAN](https://arxiv.org/abs/2010.05646v2) from [the original implementation](https://github.com/jik876/hifi-gan/blob/4769534d45265d52a904b850da5a622601885777/models.py#L75)

Adds tests:
- Smoke tests for architectures V1, V2, V3
- Check that output shapes are correct
- Check that the model is torchscriptable and scripting doesn't change the output
- Check that our code's output matches the original implementation. Here I clone the original repo inside `/tmp` and import necessary objects from inside the test function.  On test teardown I restore `PATH`, but don't remove the cloned code, so that it can be reused on subsequent runs - let me know if removing it would be a better practice

There are no quantization tests, because the model consists mainly of `Conv1d` and `ConvTransposed1d`, and they are [not supported by dynamic quantization](https://pytorch.org/docs/stable/quantization.html)

Pull Request resolved: https://github.com/pytorch/audio/pull/2860

Reviewed By: nateanl

Differential Revision: D41433416

Pulled By: sgrigory

fbshipit-source-id: f135c560df20f5138f01e3efdd182621edabb4f5

Summary:
Adds feature badges to preemphasis and deemphasis functions

Pull Request resolved: https://github.com/pytorch/audio/pull/2892

Reviewed By: carolineechen

Differential Revision: D41830782

Pulled By: hwangjeff

fbshipit-source-id: 487ce9afa8dc8fe321aa9e02cc88bb1453985d39

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/2889

Reviewed By: xiaohui-zhang

Differential Revision: D41760084

Pulled By: hwangjeff

fbshipit-source-id: d2f5253e1fae7e7aafa9fa6043c6a7045c5b33a0

Summary:
Introduces the MUSAN dataset (https://www.openslr.org/17/), which contains music, speech, and noise recordings.

Pull Request resolved: https://github.com/pytorch/audio/pull/2888

Reviewed By: xiaohui-zhang

Differential Revision: D41762164

Pulled By: hwangjeff

fbshipit-source-id: 14d5baaa4d40f065dd5d99bf7f2e0a73aa6c31a9

Summary:
Dependent on PR https://github.com/pytorch/pytorch/pull/89101

Pull Request resolved: https://github.com/pytorch/audio/pull/2853

Reviewed By: atalman, osalpekar

Differential Revision: D41737634

Pulled By: malfet

fbshipit-source-id: 715a97a2da8ef309cea78d971b47c07463495683

Summary:
This commit adds `frequency_impulse_response` function, which generates filter from desired frequency response.

[Example](https://output.circle-artifacts.com/output/job/5233fda9-dadb-4710-9389-7e8ac20a062f/artifacts/0/docs/tutorials/filter_design_tutorial.html#frequency-sampling)

Pull Request resolved: https://github.com/pytorch/audio/pull/2879

Reviewed By: hwangjeff

Differential Revision: D41767787

Pulled By: mthrok

fbshipit-source-id: 6d5e44c6390e8cf3028994a1b1de590ff3aaf6c2

Summary:
address https://github.com/pytorch/audio/issues/2885

In `_init_hubert_pretrain_model ` method which initialize the hubert pretrain models, `kaiming_normal_` should be applied on `ConvLayerBlock` instead of `LayerNorm` layer. This PR fixes it and adds more unit tests.

Pull Request resolved: https://github.com/pytorch/audio/pull/2886

Reviewed By: hwangjeff

Differential Revision: D41713801

Pulled By: nateanl

fbshipit-source-id: ed199baf7504d06bbf2d31c522ae708a75426a2d

Summary:
Adds pre-emphasis and de-emphasis functions.

Pull Request resolved: https://github.com/pytorch/audio/pull/2871

Reviewed By: carolineechen

Differential Revision: D41651097

Pulled By: hwangjeff

fbshipit-source-id: 7a3cf6ce68b6ce1b9ae315ddd8bd8ed71acccdf1

Summary:
Adds functions and transforms for speed and speed perturbation (https://www.isca-speech.org/archive/interspeech_2015/ko15_interspeech.html).

Pull Request resolved: https://github.com/pytorch/audio/pull/2829

Reviewed By: xiaohui-zhang

Differential Revision: D41285114

Pulled By: hwangjeff

fbshipit-source-id: 114740507698e01f35d4beb2c568a2479e847506

Summary:
Pull Request resolved: https://github.com/pytorch/audio/pull/2873

The original fairseq implementation had an extra layer normalization
preprocessings for large/xlarge models.

https://github.com/facebookresearch/fairseq/blob/fcca32258c8e8bcc9f9890bf4714fa2f96b6b3e1/fairseq/data/audio/hubert_dataset.py#L355-L357

This commit modifies the pre-trained model bundle to include this
preprocessing to the impacted pre-trained models listed bellow.
For the sake of keeping the interface identical to the other models,
since the additional preprocessing is rather simple, the returned
pre-trained model instance is modified ot include the preprocess,
instead of adding a method for preprocessing.

- WAV2VEC2_LARGE_LV60K
- WAV2VEC2_ASR_LARGE_LV60K_10M
- WAV2VEC2_ASR_LARGE_LV60K_100H
- WAV2VEC2_ASR_LARGE_LV60K_960H
- WAV2VEC2_XLSR53
- HUBERT_LARGE
- HUBERT_XLARGE
- HUBERT_ASR_LARGE
- HUBERT_ASR_XLARGE
- WAVLM_LARGE

Reviewed By: nateanl

Differential Revision: D41520183

fbshipit-source-id: 83d72fe692e8b9fc25df144deb4ca946fcd09615

Summary:
This commit adds `sinc_impulse_response`, which generates windowed-sinc low-pass filters for given cutoff frequencies.

Example usage:
 - [Filter Design Tutorial](https://output.circle-artifacts.com/output/job/c0085baa-5345-4aeb-bd44-448034caa9e1/artifacts/0/docs/tutorials/filter_design_tutorial.html)

Pull Request resolved: https://github.com/pytorch/audio/pull/2875

Reviewed By: carolineechen

Differential Revision: D41586631

Pulled By: mthrok

fbshipit-source-id: a9991dbe5b137b0b4679228ec37072a1da7e50bb

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/2878

Reviewed By: carolineechen

Differential Revision: D41587081

Pulled By: mthrok

fbshipit-source-id: da7f3647083a3566ce94070ce2bd30bf99e1db76

Summary:
This commit adds the tutorial for additive synthesis, using torchaudio's prototype DSP ops.

[Review here](https://output.circle-artifacts.com/output/job/3dc83322-832a-4272-9c13-df752c97b660/artifacts/0/docs/tutorials/additive_synthesis_tutorial.html)

Pull Request resolved: https://github.com/pytorch/audio/pull/2877

Reviewed By: carolineechen

Differential Revision: D41585425

Pulled By: mthrok

fbshipit-source-id: b81283b90e4779c8054fd030a1d8c3d39d676bbd

Summary:
Currently, fftconvolve only accepts the tensors for the exact same leading dimensions.
This commit loosens the restriction to allow shapes that are broadcast-able.

This makes the fftconvolve operation more efficient for cases like signal filtering where one operand (waveform) is larger than the other (filter kernel) and the same filter kernels are applied across channels and batches.

Pull Request resolved: https://github.com/pytorch/audio/pull/2874

Reviewed By: carolineechen

Differential Revision: D41581588

Pulled By: mthrok

fbshipit-source-id: c0117e11b979fb53236cc307a970a461b0e50134

Summary:
modeled after [paper](https://arxiv.org/pdf/2110.07313.pdf) and internal flow f288347302

internal comparison tests: D40080919

Pull Request resolved: https://github.com/pytorch/audio/pull/2827

Reviewed By: nateanl

Differential Revision: D41569046

Pulled By: carolineechen

fbshipit-source-id: 43c5313074af05972d93da55b2029c746b75c380

Summary:
- layer_norm in `EmformerEncoder` is set as default in emformer_hubert_model, change the type to be non-optional.
- add `aux_num_out` to emformer_hubert_model to support fine-tuning model.
- update unit tests.

Pull Request resolved: https://github.com/pytorch/audio/pull/2868

Reviewed By: carolineechen

Differential Revision: D41451311

Pulled By: nateanl

fbshipit-source-id: 5fa0f19255e4f01e001d62f8689e36f134030083

Summary:
This commits add tutorial for oscillator_bank and adsr_envelope, which will be a basis for DDSP.

 - [Review here](https://output.circle-artifacts.com/output/job/cf1d3001-88e5-418b-8cf8-ae22b4445dba/artifacts/0/docs/tutorials/oscillator_tutorial.html)

Pull Request resolved: https://github.com/pytorch/audio/pull/2862

Reviewed By: carolineechen

Differential Revision: D41559503

Pulled By: mthrok

fbshipit-source-id: 3f1689186db7d246de14f228fc2f91bf37db98cd

Summary:
Add `extend_pitch` function that can be used for augmenting fundamental frequencies with its harmonic overtones or inharmonic partials. it can be use for amplitude as well.

For example usages, see https://output.circle-artifacts.com/output/job/4ad0c29a-d75a-4244-baad-f5499f11d94b/artifacts/0/docs/tutorials/synthesis_tutorial.html

Part of https://github.com/pytorch/audio/issues/2835
Extracted from https://github.com/pytorch/audio/issues/2808

Pull Request resolved: https://github.com/pytorch/audio/pull/2863

Reviewed By: carolineechen

Differential Revision: D41543880

Pulled By: mthrok

fbshipit-source-id: 4f20e55770b0b3bee825ec07c73f9ec7cb181109

Summary:
Missing from https://github.com/pytorch/audio/issues/2848

Pull Request resolved: https://github.com/pytorch/audio/pull/2864

Reviewed By: carolineechen

Differential Revision: D41413381

Pulled By: mthrok

fbshipit-source-id: 4377ed4a59504c6ade9ee6f42938a2bc3f04fb73

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/2836

Reviewed By: carolineechen

Differential Revision: D41208630

Pulled By: nateanl

fbshipit-source-id: 625e1651f0b8a6e20876409739cf7084cb7c748b

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/2865

Reviewed By: carolineechen

Differential Revision: D41403756

Pulled By: mthrok

fbshipit-source-id: d193caa90e786f08f28e4cc2df4b4fb77aa8f592

Summary:
Adds API usage logging to MelSpectrogram and Spectrogram.

Pull Request resolved: https://github.com/pytorch/audio/pull/2861

Reviewed By: carolineechen

Differential Revision: D41384080

Pulled By: hwangjeff

fbshipit-source-id: caf4b0fa6e4cc3954384bfdd08a183b90d07d974

Summary:
Add adsr_envelope op, which generates ADSR envelope

* Supports generation of the envelope on GPU
* Supports optional Hold
* Supports polynomial decay

<image src='https://download.pytorch.org/torchaudio/doc-assets/adsr_examples.png'>

Pull Request resolved: https://github.com/pytorch/audio/pull/2859

Reviewed By: nateanl

Differential Revision: D41379601

Pulled By: mthrok

fbshipit-source-id: 3717a6e0360d2a24913c2a836c57c5edec1d7b31

Summary:
This code was added by
https://github.com/pytorch/audio/commit/4d0095a528412cfec2a549204fc01d9ebb15df7a

Seems that the original code had a typo?

Pull Request resolved: https://github.com/pytorch/audio/pull/2858

Test Plan:
```
// the import of `mustc` now succeeds, previously crashed
python examples/asr/emformer_rnnt/global_stats.py --model-type librispeech --dataset-path /home/vasiliy/local/librispeech/
```

Reviewed By: carolineechen

Differential Revision: D41355663

Pulled By: nateanl

fbshipit-source-id: 92507e529d41b984b9dd400ad24a55d130372b7d

Summary:
This commit adds `oscillator_bank` op, which is the core of (differential) digital signal processing ops.
The implementation itself is pretty simple, sum instantaneous frequencies, take sin and multiply with amplitudes.

Following the magenta implementation, amplitudes for frequency range outside of [-Nyquist, Nyquist] \
are suppressed.

The differentiability is tested within frequency range of [- Nyquist, Nyquist], and amplitude range of [-5, 5], which should be enough.

For example usages:
 - https://output.circle-artifacts.com/output/job/129f3e21-41ce-406b-bc6b-833efb3c3141/artifacts/0/docs/tutorials/oscillator_tutorial.html
 - https://output.circle-artifacts.com/output/job/129f3e21-41ce-406b-bc6b-833efb3c3141/artifacts/0/docs/tutorials/synthesis_tutorial.html

Part of https://github.com/pytorch/audio/issues/2835
Extracted from https://github.com/pytorch/audio/issues/2808

Pull Request resolved: https://github.com/pytorch/audio/pull/2848

Reviewed By: carolineechen

Differential Revision: D41353075

Pulled By: mthrok

fbshipit-source-id: 80e60772fb555760f2396f7df40458803c280225

Summary:
address https://github.com/pytorch/audio/issues/2847

In mixed precision training, the dtype of `mask_embedding` is **not** converted to fp16 automatically. This PR addresses the issue by changing the dtype of `mask_embedding` to `x` to enable mixed precision training.

Pull Request resolved: https://github.com/pytorch/audio/pull/2854

Reviewed By: carolineechen

Differential Revision: D41343486

Pulled By: nateanl

fbshipit-source-id: 4a5cbb429ff8ba5d3c439a3d5acb5094f66bf705

Summary:
- `_get_fileids_paths` in `LibriLightLimited` dataset was changed dataset in https://github.com/pytorch/audio/issues/2653, the absolute path becomes relative paths. This PR fixes the usage in hubert fine-tuning recipe to get correct audio paths.
- model options should be `hubert_pretrain_large` and `hubert_pretrain_xlarge` instead of `hubert_large` and `hubert_xlarge`.
- The input dimension of CTC linear layer varies depending on the model architecture, update it in lightning module.

cc simpleoier

Pull Request resolved: https://github.com/pytorch/audio/pull/2851

Reviewed By: carolineechen

Differential Revision: D41327998

Pulled By: nateanl

fbshipit-source-id: f92248ee84ec860b4e4dbef880c5794b338e1e2d

Summary:
Closes T136364380, follow-up to https://github.com/pytorch/audio/issues/2822

- Added "base", "base+", and "large" bundles for WavLM
- Expanded `wav2vec2_pipeline_test.py` to include the new bundles
- Added the new bundles to docs in `pipelines.rst`

Pull Request resolved: https://github.com/pytorch/audio/pull/2833

Reviewed By: nateanl

Differential Revision: D41194796

Pulled By: sgrigory

fbshipit-source-id: bf8e96c05b6a81ac5c5a014c46adeeac12685328

Summary:
Closes T137506059

Replaces functional multi-head attention in `WavLMSelfAttention` with a module `torch.nn.MultiheadAttention`. The reason is that the latter uses native CPU/CUDA implementation ([BetterTransfomer](https://pytorch.org/blog/a-better-transformer-for-fast-transformer-encoder-inference/)) under certain conditions, and can achieve significant speedup. It also simplifies the code in `WavLMSelfAttention`

Note: the definition of `bias` parameter in `WavLMSelfAttention.forward` has changed slightly, because in `torch.nn.MultiheadAttention` there is no parameter controlling presence of bias for projections of `k`, `v`, and `q` independently. In WavLM we only use `bias=True`, so it won't have any effect for users of WavLM or tests

Pull Request resolved: https://github.com/pytorch/audio/pull/2842

Reviewed By: nateanl

Differential Revision: D41186166

Pulled By: sgrigory

fbshipit-source-id: e791c68106ad89f96c1abf046de699cb8ec7b595

Summary:
* Add the new official torchaudio logo to documentation/README.
* Add a page for download logo.

https://output.circle-artifacts.com/output/job/e9eb1292-7c10-4fef-adc3-ad568802aa59/artifacts/0/docs/index.html

<img width="1068" alt="Screen Shot 2022-11-14 at 10 30 27 AM" src="https://user-images.githubusercontent.com/855818/201738349-9e248f15-dce2-4931-9066-aa898a53d6ad.png">

https://output.circle-artifacts.com/output/job/e9eb1292-7c10-4fef-adc3-ad568802aa59/artifacts/0/docs/logo.html

<img width="617" alt="Screen Shot 2022-11-14 at 10 30 47 AM" src="https://user-images.githubusercontent.com/855818/201738420-ad0fda2f-f310-4802-851c-bbdf6c84c045.png">

Pull Request resolved: https://github.com/pytorch/audio/pull/2802

Reviewed By: carolineechen

Differential Revision: D41295277

Pulled By: mthrok

fbshipit-source-id: 6615d00799c9611f875e8485459d800e350b3486

Summary:
Removing LTS mention and packages from README as it is discontinued.

Pull Request resolved: https://github.com/pytorch/audio/pull/2844

Reviewed By: hwangjeff, xiaohui-zhang

Differential Revision: D41200886

Pulled By: mthrok

fbshipit-source-id: 0da0afe68df51826075ce945cf0cf1de901e1c8f

Summary:
follow up to https://github.com/pytorch/audio/issues/2823
- move bark spectrogram to prototype
- decrease autograd test tolerance (passing on circle ci)
- add diagram for bark fbanks

cc jdariasl

Pull Request resolved: https://github.com/pytorch/audio/pull/2843

Reviewed By: nateanl

Differential Revision: D41199522

Pulled By: carolineechen

fbshipit-source-id: 8e6c2e20fb7b14f39477683b3c6ed8356359a213

Summary:
address https://github.com/pytorch/audio/issues/2845

Pull Request resolved: https://github.com/pytorch/audio/pull/2846

Reviewed By: carolineechen

Differential Revision: D41251624

Pulled By: nateanl

fbshipit-source-id: 1a363d2314d6a452f35c109b9730da64ada5a2fd

Summary:
Added missed build workflows for MacOS and Linux:

- [x] Linux conda
- [x] MacOS conda

This does not change the existing builds/uploads in CircleCI, and should not break any existing jobs/workflows. This is just to add back workflows for the MacOS and Linux conda builds with Nova.

We will create a workflow (most likely in test-infra) that does this comparison between the binaries to ensure there is parity between the binaries before we start uploading with Nova.

Pull Request resolved: https://github.com/pytorch/audio/pull/2800

Reviewed By: osalpekar

Differential Revision: D41181467

Pulled By: DanilBaibak

fbshipit-source-id: a5c5d4dcfdd778b4045203f6016c20fb42daa01b

Summary:
Currently `discard_before_pts=-1` is used to indicate no AVFrame should be skipped. It was reported that some corrupted video can have constant negative pts value.

It is technically UB for such corrupted data, but still all the AVFrame should be decoded as long as `seek` is not used.

This commit changes the decoder so that it processes AVFrame if `discard_before_pts==-1` disregard of AVFrame::pts value.

Pull Request resolved: https://github.com/pytorch/audio/pull/2841

Reviewed By: hwangjeff

Differential Revision: D41174442

Pulled By: mthrok

fbshipit-source-id: e9d2fab4b0e2bc47146eda8e1dd377a74c087590

Summary:
Adding Nova Reusable Workflow for M1 Wheels Build. Once this has been running well for a while, we can replace the old `build-m1-binaries.yml` workflow.

Pull Request resolved: https://github.com/pytorch/audio/pull/2839

Reviewed By: DanilBaibak

Differential Revision: D41195316

Pulled By: osalpekar

fbshipit-source-id: f3754043f384b1645e5fcfaebf465f6839f72461

Summary:
Adding Nova Reusable Workflow for M1 Conda Build. Once this has been running well for a while, we can replace the old `build-m1-binaries.yml` workflow.

Pull Request resolved: https://github.com/pytorch/audio/pull/2840

Reviewed By: DanilBaibak

Differential Revision: D41195298

Pulled By: osalpekar

fbshipit-source-id: 14591b96e998aa43fa57e8e5b0b09d0ce4f4092e

Summary:
I have added BarkScale transform, which can transform a regular Spectrogram into a BarkSpectrograms similar to MelScale. ahmed-fau opened this requirement in December 2021 with the number (https://github.com/pytorch/audio/issues/2103). The new functionality includes three different well-known approximations of the Bark scale.

Pull Request resolved: https://github.com/pytorch/audio/pull/2823

Reviewed By: nateanl

Differential Revision: D41162100

Pulled By: carolineechen

fbshipit-source-id: b2670c4972e49c9ef424da5d5982576f7a4df831

Summary:
internal comparison tests: D40080919

follow up PR for pretrained models https://github.com/pytorch/audio/issues/2827

Pull Request resolved: https://github.com/pytorch/audio/pull/2826

Reviewed By: nateanl

Differential Revision: D41160061

Pulled By: carolineechen

fbshipit-source-id: f3c478b28c235af53d1d8e21b573c53684a63ac4

Summary:
Closes T136364380

Added [WavLM Model](https://github.com/microsoft/UniSpeech/tree/main/WavLM):
- Added `WavLMSelfAttention` class (from [original implementation](https://github.com/microsoft/UniSpeech/blob/2e9dde8bf815a5f5fd958e3435e5641f59f96928/WavLM/modules.py)) and adjusted existing Encoder and Transformer classes to be compatible with it
- Added factory functions `wavlm_model`, `wavlm_base`, `wavlm_large` to `models/wav2vec2/model.py`
- Added bundles for base and large models to pipelines. **TODO**: pre-trained model weights are not yet uploaded to `download.pytorch.org`, permissions not granted yet.

## Tests
- Expanded HuggingFace integration tests to cover WavLM. For there tests, added JSON configs for base and large models from HF ([base](https://huggingface.co/microsoft/wavlm-base/blob/main/config.json), [large](https://huggingface.co/microsoft/wavlm-large/blob/main/config.json)) into test assets
- Expanded TorchScript and quantization tests to cover WavLM

## Comments
There are a few workarounds I had to introduce:
- Quantization tests for WavLM were breaking down at [`torch.cat`](https://github.com/pytorch/audio/pull/2822/files#diff-6f1486901c94320ec0610a460dc674638fab9d104a61564ff7b59353a8b8547cR466) ~~until I excluded the arguments of `torch.cat` from quantization [here](https://github.com/pytorch/audio/pull/2822/files#diff-6f1486901c94320ec0610a460dc674638fab9d104a61564ff7b59353a8b8547cR368-R369). I haven't found a better way to fix it, let me know if there is one~~ The reason for this seems to be that quantization replaces `.bias` and `.weight` attributes of a `Linear` module with methods. Since we are using weights and biases directly, the code was break. The final solution suggested by nateanl was to define attention weights and biases directly in `WavLMSelfAttention`, skipping the `Linear` layers
- ~~WavLM uses position embedding in the first layer of encoder, but not in the subsequent ones.  So [UniSpeech](https://github.com/microsoft/UniSpeech/blob/2e9dde8bf815a5f5fd958e3435e5641f59f96928/WavLM/modules.py#L342) and [HF](https://github.com/huggingface/transformers/blob/b047472650cba259621549ac27b18fd2066ce18e/src/transformers/models/wavlm/modeling_wavlm.py#L441-L442) implementations only create this embedding module in the layers where it's used. However, we can't do this here because it breaks TorchScript. So as a solution I add a dummy `Identity` module to `WavLMSelfAttention` when the actual embedding is not needed: [here](https://github.com/pytorch/audio/pull/2822/files#diff-6f1486901c94320ec0610a460dc674638fab9d104a61564ff7b59353a8b8547cR361-R368).~~ Thanks nateanl for resolving this!
- I had to add dummy `position_bias` and `key_padding_mask` arguments to `SelfAttention.forward` to make TorchScript tests pass. Since both `SelfAttention` and `WavLMSelfAttention` are called from `EncoderLayer`, they need to have compatible signatures. Having a variable number of arguments with `**kwargs` or checking object class doesn't seem to work with TorchScript, so I instead made both types of attention accept `position_bias` and `key_padding_mask` arguments.

Nit: do we still need to specify `__all__` if there are no wildcard imports in `__init__.py`, e.g. in `torchaudio/models/__init__.py`?

Pull Request resolved: https://github.com/pytorch/audio/pull/2822

Reviewed By: nateanl

Differential Revision: D41121855

Pulled By: sgrigory

fbshipit-source-id: 9f4f787e5810010de4e74cb704063a26c66767d7