SUMMARY
- The front-end module processes speech data into
observation vectors which represent events in the probability
space.
- Standard DSP algorithms (FFT, LP, FBA) are specialized
to the speech problem by adding expert physiological knowledge of
the vocal articulatory and auditory perceptual systems.
- Although the classification experiments failed,
the features do make sense. A comparison of our coefficients with
other state-of-the-art systems shows a similar plot.