Machine learning lie detector
Setup
I will not publish the values of the parameters needed to run this code because these values are sort of subjectives and because keeping them for myself will keep me a little ahead (presumably. Instead, I will publish the empty configuration file: locals.cfg.template.
Thoughts log:
- Download you tubes "interviews" of all kinds
- Extract audio
- Heuristic way to select audios without background music
- Extract from audio short segments, that is, words or voice sounds
- Extract features from segments
- Normalize
- Make clusters
- Decision point
If interested, contact me through github.com. Comments and suggestions would be greatly appreciated!