The final batch of utterances have been uploaded.
It seems the additional audio files (the previous, and current) are mp3 encoded, while the original audio files are wav encoded. Reading mp3 files with a popular package like librosa is slow compared to reading wav files and therefore creates a bottleneck during model training. I want to believe we're allowed to convert the files to wav format?
Yes, you can convert mp3 to wav.
Thanks, is it just a compilation of the two first datasets, or we have new utterances or cleaning of the two first datasets?
I think it is news because the number of samples are differents
oh, looks like. Didn't read the word "batch". thanks
It looks like the new nlp_keywords_29Oct2020.zip sound files are in exactly the same format as AdditionalUtterances.zip. Please confirm.