Audio decoding format issues

[CarolinePascal]

🐛 Describe the bug

Hello torchcodec team !

Thank you for the last 0.3 release and the long-awaited support for ✨audio decoding✨ ! For now, AudioDecoder works like a charm, but I have a few remarks that would make it even more pleasant to use :

1. AudioSamples shape format

As for now, audio data stored in AudioSamples is stored in a tensor of shape [num_channels, num_samples] whereas most audio library (including sounddevice and soundfile) opt for a [num_samples, num_channels] shape. It's just a transposition, and it would be nice to match the reference shape !

2. Output format when start_seconds and stop_seconds are identical

When start_seconds and stop_seconds are set to the same value the audio data stored in AudioSamples "loses" the num_channels dimension, which is set to 0. This actually led to a bunch of issues on my side, and I think it would be nice to either raise the Invalid start seconds issue when start_seconds equals stop_seconds or keep the first dimension equals to the number of channels.

Thank you again for your work on audio decoding, and for taking my remarks into account !

Versions

PyTorch version: 2.7.0
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 15.4.1 (arm64)
GCC version: Could not collect
Clang version: 16.0.0 (clang-1600.0.26.6)
CMake version: version 4.0.0
Libc version: N/A

Python version: 3.12.2 | packaged by conda-forge | (main, Feb 16 2024, 20:54:21) [Clang 16.0.6 ] (64-bit runtime)
Python platform: macOS-15.4.1-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Apple M4 Pro

Versions of relevant libraries:
[pip3] mypy-extensions==1.0.0
[pip3] numpy==2.2.4
[pip3] torch==2.7.0
[pip3] torchaudio==2.7.0
[pip3] torchcodec==0.3.0
[pip3] torchvision==0.22.0
[conda] libopenvino-pytorch-frontend 2024.4.0 h5833ebf_2 conda-forge
[conda] numpy 2.2.4 pypi_0 pypi
[conda] torch 2.7.0 pypi_0 pypi
[conda] torchaudio 2.7.0 pypi_0 pypi
[conda] torchcodec 0.3.0 pypi_0 pypi

[NicolasHug]

Thank you for the report @CarolinePascal !

I've opened #663 to keep track of the sample shape feature request.

I also opened #664 which should fix the shape bug you reported!