Speech recognition

Things you can do easily to improve speech recognition.

Some part of this Section is still under construction.

.dic and .jconf files.

  • Adding words
  • Modifying parameters
  • More things in devdoc

MMDAgent includes full version of Julius.

Basic behavior

RECOG_EVENT_START, RECOG_EVENT_STOP

How to capture words in FST

RECOG_EVENT_STOP

AND, OR condition

Add words

Define task-specific dictionary in your content to add task-specific words.

Multiple user dictionary by message.

Change setting

If per-task setting file exists, it will be loaded after system jconf, overriding system default. Default level threshold, beam width, other parameters.

Full customization

The same model as Julius can be used. All the settings and models that can be used in Julius is also applicable to MMDAgent. Put models and set the jconf file as …


Last modified January 19, 2020: Major update and re-construction, make all links to rel. (5fcaabd)