BY: JOSH TABOR Applying Multilayer Perceptron Artificial Neural Networks to Recognizing Piano Keystrokes
The Project Create an MLP ANN to correctly identify which piano keys are pushed based on their FFT coefficients Test ANN at different noise levels and maybe on different pianos
The Plan Collect data (Middle C – Tenor C) Keys to be used
The Plan (continued) Antialiasing Filter Downsample Sampled at 44.1Khz Highest f= 523Hz Downsample to 1200Hz Saves processing time Breakup signal
The Plan (continued) Take FFT Average windows Label Develop ANN Test ANN
Expected Results Expect it to work fairly well (90% classification rate) FFT cleaner than expected Performance degrades with SNR decrease