As part o f my PhD work it was necessary to create a few routines and programs to analise voices and classify them as "normal" or "pathological".
The (library) routines are listed here.
The programs (c sources for creating executables) are:
crop - Downsample a 50000 sample per second file to 25000 samples per second, saving only the first second of the file.
voice_par - Calculate various parameter vectors (LPC, Cepstrum, delta cepstrum, weighted cepstrum) for a list of voice files.
create_cb - create codebook creates a codebook dictionary of representative vectors for a group of voice files.
vq - Vector Quantization calculates the distortion of a file in relation to a codebook.