Next: 5.15 Summary
Up: 5 Speech Input/Output
Previous: 5.13 Copying and Coding using HCOPY
The redesign of the HTK front-end in version 2.0 has introduced
a number of differences in parameter encoding. The main
changes are
- Source waveform zero mean processing is now performed on a frame-by-frame
basis.
- Delta coefficients use a modified form of regression rather than
simple differences at the start and end of the utterance.
- Energy scaling is no longer applied to the zero'th MFCC coefficient.
If a parameter encoding is required which is as close as possible
to the version 1.5 encoding, then the compatibility configuration
variable V1COMPAT should be set to true.
Note also in this context that the default values for the various
configuration values have been chosen to be consistent with the
defaults or recommended practice for version 1.5.
Next: 5.15 Summary
Up: 5 Speech Input/Output
Previous: 5.13 Copying and Coding using HCOPY
ECRL HTK_V2.1: email [email protected]