test and val dataset included in train in speech_command_classification_with_torchaudio_tutorial.py

[algat]

In speech_command_classification_with_torchaudio_tutorial.py, the test and validation datasets are included in the training dataset.

When doing:

tutorials/intermediate_source/speech_command_classification_with_torchaudio_tutorial.py

Lines 95 to 96 in bb3523f

	train_set = SubsetSC("training")
	test_set = SubsetSC("testing")

And checking for intersection of speakers:

train_speaker_ids = set([ s[3] for s in train_set])
test_speaker_ids = set([ s[3] for s in test_set])
intersection_of_speakers = train_speaker_ids & test_speaker_ids
print(len(train_speaker_ids))
print(len(test_speaker_ids))
print(len(intersection))

It gives

2618
250
250

Hence the 250 speakers in test are in the training dataset. Same for thing for validation set.