HMM Toolkit STK

A set of tools for training and recognition of HMM's. Probably the most interesting in this distribution is SERest - a tool for embedded training of HMM's with supporting scripts. Key features of SERest include re-estimation of linear transformations (MLLT, LDA, HLDA) within the training process, and use of recognition networks for the training.


Lukáš Burget, Ondřej Glembek, Petr Schwarz, Martin Karafiát


The current version number is 2006_12_18
Binary distribution for Linux (I386), including some supporting scripts is available in stk_bin_linux-i386_2006_12_18.tgz
Source distribution is available in stk_src_2006_12_18.tgz


  • Example of estimating HLDA transformation for HMM based phoneme recognizer.
  • Example of re-training phoneme recognizer using MMI and MPE criteria.
  • Example of re-training GMM based language recognition system using MMI criterion.


This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.