Using HTK & Julius
HTK (Hidden Markov Toolkit) is a portable toolkit for building and manipulating the statistical models used to represent sound in Speech Recognition (these are called hidden Markov models). Julius is large vocabulary continuous speech recognition engine.
This tutorial runs through the steps to adapt the VoxForge Speaker
Independent Acoustic Model to your voice using the HTK
toolkit, thus increasing its recognition
accuracy for your voice, which can then be used with the Julius Speech Recognition Engine.
Note: This Tutorial uses HTK release 3.2.1 because there are problems with adaptation using HTK release 3.3 (see Ticket #41 for details).
Julius works with models trained with either HTK release 3.2.1 or HTK release 3.3.
Table of ContentsDownload HTK 3.2.1 & SoX
Step 1 - Prepare Data
Step 2 - Coding the Data
Step 3 - Creating the Transcription Files
Step 4 - Realigning the Training Data
Step 5 - Generating the Transforms
Live Testing with Julian