Undocumented Aspects of MFCC Computation by AudioLocalMeasurements

Question

There are a number of undocumented procedures and parameters underpinning the computation of MFCC vectors by AudioLocalMeasurements including

1. the size and form of the filter bank,
2. the liftering and pre-emphasis formulae applied and their parameters,
3. the form and parameters of the Fourier transformations used,
4. the nature of the normalization of the audio data, etc.

I'm offering a bounty to reproduce the MFCC vectors by AudioLocalMeasurements using Mathematica primitives (FourierDCT, PeriodogramArray, etc.) so that one can understand and make use of the MFCC output.