11.1 Making a Julius configuration file (.jconf)

Problem

Read this section if you do not know how to make a Julius configuration file (.jconf file) and what options should be set.

Solution

Julius includes many optional parameters that can be set by the user. It is complicated to designate options in Julius every time using command lines. A series of optional parameters can be set in a text file, thus simplifying option inputs in Julius. This text file is called a .jconf file. The options used in Julius are summarized in http://julius.sourceforge.jp/juliusbook/ja/desc option.html. All options can be described in text in a .jconf file. Important reminders when connecting Julius to SpeechRecognitionClient (or SpeechRecognitionSMNClient ) are summarized. The minimum setting items are

-notypecheck
-plugindir /usr/local/lib/julius_plugin
-input mfcnet
-gprune add_mask_to_safe
-gram grammar
-h hmmdefs
-hlist allTriphones

-notypecheck is an essential option. HARK uses an expanded acoustic parameter structure and is not supported by type check of the Julius default. Therefore, it is essential to add -notypecheck. When omitting this option, Julius detects a type error in the type check of features and does not recognize sounds.

For -plugindir, designate the path in which function enhancement plug-in files such as mfcnet are saved. This path must be designated before -input mfcnet and -gprune add masks. All enhancement plugs in files present in the plug-in path are read.

-input mfcnet is an option to recognize features received from SpeechRecognitionClient (or SpeechRecognitionSMNClient ). Designate this option to support missing feature masks. Select the Gaussianpruning algorithm to support -gprune. Since calculations are performed while supporting the missing feature masks, the option name is different from that for normal Julius. Select from $\{ add\_ mask\_ to\_ safe$ $||add\_ mask\_ to\_ heu$ $|| add\_ mask\_ to\_ beam$ $|| add\_ mask\_ to\_ none\}$ , which correspond to $\{ safe$ $||heuristic$ $||beam||none\}$ of normal Julius, respectively.

In addition, it is necessary to designate a language model and an acoustic model, the same as for normal speech recognition. Designate a grammar file in -gram, a definition file in -h HMM and an HMMList file in -hlist.