Reverse-engineer the SensMe classifier #21

New Issue

lucidiot · 2020-01-22T21:40:31Z

lucidiot commented

2020-01-22 21:40:31 +00:00

First, we can try to just use the song corpora offered by SOTD (mastodon bot, text file, script with json output at ~severak/bin/sotds), just classify it, then get a copy of the SENSME.DAT SQLite database to see how the songs were classified.

Then, we can try to compute some criteria (tempo/mood detection) and find out how songs are clusterized by the SensMe app; this can then be confirmed by generating random MP3s matching specific frequencies, tempos, etc. and experimenting with SensMe.

This would generate the most scientific-looking part of this PSP research project, with so many graphs!

First, we can try to just use the song corpora offered by SOTD ([mastodon bot](https://tiny.tilde.website/@sotd), [text file](tilde.town/~severak/allsotds.txt), script with json output at `~severak/bin/sotds`), just classify it, then get a copy of the `SENSME.DAT` SQLite database to see how the songs were classified. Then, we can try to compute some criteria (tempo/mood detection) and find out how songs are clusterized by the SensMe app; this can then be confirmed by generating random MP3s matching specific frequencies, tempos, etc. and experimenting with SensMe. This would generate the most scientific-looking part of this PSP research project, with so many graphs!

❤️ 1

lucidiot added the

help wanted

requires-hardware

labels 2020-01-22 21:40:31 +00:00

Sign in to join this conversation.