TODO:

• Model/Algorithm to determine signal presence in incoming audio (100 or 200 pixels ahead)
• Noise generator class to apply realistic noise patterns
• Global class to log model training parameters and results to json file for visualizatoin and db storage
• Log batch size to results
• Collect global variables for all training and fft-flows. eg. tempo interval, cropping, image size etc
• fix json serialize model config in categorical data log
• add type checking to logging class properties