search

bmcfee / ismir2017_chords

ISMIR 2017: structured training for large vocab chord recognition
BSD 2-Clause "Simplified" License
48 stars 3 forks source link

Confused by the "synthetic data" in your Data Pump code #4

Closed cloudscapes closed 4 years ago

cloudscapes commented 4 years ago

Hi dear Brian,

I'm a bit confused by the " Synthetic Data" in your 01 - Data pump.ipynb file. If I'm not mistaken, in your 00 - Augmentation.ipynb file you have a separate directory for the audio files and another one for the corresponding annotations. In the 01 - Data pump.ipynb file you build two separate pumps, one for the original audio files and their annotations and one for the augmented audio files and their annotations and transform and save each of those two separately. But after that, there is a code snippet that deals with synthetic data that lies in a completely different folder!

Are these " synthetic data" midi-synthesized audio files ?? I assume I won't need them and having the audio files and their annotations is enough to keep me going?

Thanks in advance and with best regards,

H.F

bmcfee commented 4 years ago

But after that, there is a code snippet that deals with synthetic data that lies in a completely different folder!

Are these " synthetic data" midi-synthesized audio files ?? I assume I won't need them and having the audio files and their annotations is enough to keep me going?

Good catch! This is a hold-over from some earlier experiments with synthetic chord templates (not even midi files) that we were trying to use to train the decoder. This didn't make it into the final version of the paper, and it's safe to ignore this bit.

cloudscapes commented 4 years ago

Thank you very much!