[wip] Decode WAV with C++ backend

[Dan-Flores]

To review:

• First, see the changes to the public AudioDecoder class.
  • This implementation changes the _audio_decoder.py Python file. To completely contain changes in C++ requires more complex changes, ex. implementing a Decoder class that SingleStreamDecoder and WavDecoder implement.
• Read _is_uncompressed_wav, and read a function it dispatches to, like get_wav_metadata_from_file.
  • Skim through the input type handler classes, WavFileReader and WavTensorReader
• Read WavDecoder::WavDecoder init implementation, follow it to parseHeader.
• Wonder at convertSamplesToFloat
  • Using tensor operations here reduces the performance gains. This implementation does the conversion in a single pass.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/meta-pytorch/torchcodec/1221

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Lint/Mac jobs intermittently fail

❌ 1 New Failure

As of commit c01d370 with merge base 377c638 ():

NEW FAILURE - The following job has failed:

• Lint / mypy (3.12) (gh)
  Process completed with exit code 1.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Dan-Flores]

The AudioDecoder exposes the input audio's metadata to the user. Using a C++ backend without FFmpeg, we pass over the information necessary to create AudioStreamMetadata from the C++ side. This implementation uses JSON, but its possible to pass each field individually and construct an AudioStreamMetadata object here as well.

[Dan-Flores]

self._wav_source is only populated if WAV decoding was successful.

[Dan-Flores]

Once we find the RIFF and WAVE signatures, we can look for the fmt chunk which contains metadata, and the data chunk which contains the actual samples. We find the data chunk now to store dataSize and dataOffset, which will be needed later for decoding.