The Neural Mixer

This was originally a project for an ECE elective course I had taken. This is an implementation of a fully automated "Neural Mixer" which consumes raw, unleveled, or EQ-ed audio that you wish to combine. The network is fuzzy logic, which is really just an RBF with a GUI ( I WILL ARGUE THIS ALL DAY WITH FIS SUPPORTERS! ). Equalization is done by gaining the spectrum of some finite sample, over the entire file duration, and comparing it with a known standard, that a human (ME) deems is correct. This is the training procedure. A similar implementation is used for adjusting amplitudes, although it is a power measurement, not a spectral...

Here are all of the audio files for the FIS Neural Mixer I have created:

Data Set 1:

1A - Raw audio tracks, unmixed

1A - Audio tracks, mixed with Volume FIS only

1A - Audio tracks, mixed with both the Volume FIS and Spectral Gain FIS

1B - Raw audio tracks, unmixed

1B - Audio tracks, mixed with Volume FIS only

1B - Audio tracks, mixed with both the Volume FIS and Spectral Gain FIS

Data Set 2:

2A - Raw audio tracks, unmixed

2A - Audio tracks, mixed with Volume FIS only

2A - Audio tracks, mixed with both the Volume FIS and Spectral Gain FIS