v0.9 release checklist

[mthrok]

Note: This is not exhaustive list.

Before branch cut

Deprecation

• Set removal release version #1535 Update deprecation warnings so that instead of saying "feature release", it says specific release version. F.resample, T.MelScale
• Deprecate kaldi resample #1533 Deprecate torchaudio.compliance.kaldi.resample_waveform.
• [BC-Breaking] Remove deprecated arguments from CommonVoice #1534 Remove deprecated arguments from CommonVoice

Documentation

• Add autograd support to related functionals/transforms.
• [Cherry-picked 0.9] Update docs #1550 Overall documentation update. We need to review the documentation and fix rendering / update docstrings that are not easy to understand.
  • Fix DeepSpeech link rendering https://pytorch.org/audio/master/models.html#deepspeech
  • Sound file backend: Note of info[code, render], load[code, render] and save function.
  • wav2vec2 model: Add function index to the category on the right side.

Independent of branch cut

Example

• Add speech recognition C++ example #1538 C++ speech recognition example

Tutorial

• Update audio manipulation tutorial  tutorials#1566
  • Update tutorial about HTK support.
    There are comments about limited HTK compatibility (Mel Filter Bank and MelSpectrogram), but this was resolved in enable mel_scale option #593. We should update the tutorial. The source code is found here and here.
  • Add resampling tutorial (as outlined in Resampling #1487)

Optional

• Resolve Simplify axis value checks #1501
• Resolve [ROCm] add rocm wheels #1477 @malfet

@carolineechen @cpuhrsch @vincentqb

[mthrok]

@carolineechen I think resample is better suited under Utility category.

https://pytorch.org/audio/master/functional.html#resample

Also Note: should be a new section like Args: so that it gets attention in documentation.

audio/torchaudio/transforms.py

Line 677 in fad19fa

Note: If resampling on waveforms of higher precision than float32, there may be a small loss of precision

Example:
code -> rendered