transfer correlate characteristics between alcoholic and sober EEG signals w/ DDSP. Done for BME 512
- DDSP provides notebooks for training an autoencoder and timbre transfer -- not included here.
- Extremely janky preprocessing creates mostly-zero spectrogram
- this neuters spectrogram loss since leaving signal unchanged is "mostly right"
- could either fix preprocessing or attempt to "class-balance" the spectrogram loss
- this neuters spectrogram loss since leaving signal unchanged is "mostly right"
- Modules used were fairly limited and EEG-nonspecific