mistake in torchaudio data augmentation tutorial

[Moadab-AI]

In the "adding background noise" section of the tutorial file :
https://github.com/pytorch/tutorials/blob/master/beginner_source/audio_data_augmentation_tutorial.py

first of all you are calculating RMS but calling it power:

speech_power = speech.norm(p=2)
noise_power = noise.norm(p=2)

and then since you're dealing with RMS, the snr_db should be divided by 20 and not 10:
snr = math.exp(snr_db / 10)

also for a "standard" SNr value you'd have to use to log in base 10 as well i suppose.

[svekars]

@mthrok can you take a look?

[mthrok]

Hi @Moadab-AI

Thanks for the issue. I think this has been fixed in torchaudio side pytorch/audio#2285, and the change was ported in tutorial site #1939, which happened in June.

Can you confirm that https://pytorch.org/tutorials/beginner/audio_data_augmentation_tutorial.html#adding-background-noise looks okay now?

Separately, I just realized that it has not been fixed on the next section; https://pytorch.org/tutorials/beginner/audio_data_augmentation_tutorial.html#simulating-a-phone-recoding

[svekars]

This tutorial was moved. If this is still an issue, please raise in the pytorch/audio repo.