Audio Processing Sandbox (Ongoing)

Personal Jupyter notebooks and Python scripts used to experiment on my guitar recordings. Mono to stereo conversion for a moving source, including frequency dependent path loss, doppler effect, and rudimentary considerations for the low pass filter from the geometry of the ear. Adaptive noise cancellation using NLMS filtering, mimicking filtering done in modern headphones without secondary path considerations. Sustained note extraction from a given audio file. The criteria for classifying a note as sustained includes the slope and magnitude of RMS energy over time and the stability of the estimated pitch by YIN. Timbre replication given a reference recording, the same recording with some change in timbre, and a separate recording on which the filter is applied. Implemented with mean FFT ratios of sustained notes. I'm currently working on source separation using Non-negative Matrix Factorization with constraints on sparsity and harmonicity, and a custom tunable amp model GUI with an initial nonlinearity (tanh) for distortion, EQ bandpass filters, a cut-and-paste filter to get a desired timbre, an output nonlinearity for compression, and added reverb (echo and decay).