Speech commands loading broken

[neworderofjamie]

When using norse.datasets.speech_commands.prepare_dataset, the file lists generated at https://github.com/norse/norse/blob/main/norse/dataset/speech_commands.py#L57 and https://github.com/norse/norse/blob/main/norse/dataset/speech_commands.py#L63 start with ./ e.g. ./SpeechCommands/speech_commands_v0.02/six/c2d15ea5_nohash_0.wav but those produced by the walkers at https://github.com/norse/norse/blob/main/norse/dataset/speech_commands.py#L68, https://github.com/norse/norse/blob/main/norse/dataset/speech_commands.py#L74 and https://github.com/norse/norse/blob/main/norse/dataset/speech_commands.py#L77 do not so everything ends up in training set. Afraid I don't know enough about torchaudio to know the root cause of this, otherwise I'd make a PR.

Finally, pretty sure it should be test_filelist and test_sc._walker at https://github.com/norse/norse/blob/main/norse/dataset/speech_commands.py#L77

System:
Ubuntu 18.04
Python 3.8
TorchAudio 2.1.2+cu121
Torch 2.1.2+cu121

[Jegp]

Thanks for the issue! Unfortunately, I'm a little time-restricted currently. If I recall correctly, the original implementor was @cpehle. Do you have a minute to check this out?