Julius (software)

from Wikipedia, the free encyclopedia
Julius
Basic data

Maintainer Lee Akinobu ( Nagoya Institute of Technology )
Current  version 4.5
( January 2, 2019 )
operating system Unix-like ( GNU / Linux , BSD etc.), Windows (via Cygwin )
programming language C.
category voice recognition
License own license
German speaking Yes
julius.sourceforge.jp

Julius is a voice recognition - Engine , as free software was implemented. It belongs to the LVCSR (Large Vocabulary Continuous Speech Recognition) genre , which understands uninterrupted language flows and can deal with large vocabulary. The program is command line based, but there is also a graphical front end called Simon .

Technical background

Julius is based on Hidden Markov Models and N-gram technology, which divides the incoming signal into groups of several words (in the case of Julius two or three, i.e. bigrams and trigrams), the probability of which is calculated on the basis of statistical data. In order to be independent of the language model used, the program has a modular structure; it supports HTK and the CMU-Cam SLM toolkit , among other things . The program runs almost in real time on modern PCs. The word accuracy depends on the language model used.

Development history

The project was started in Japan in 1997 , where it is used at various universities and institutes for research in computational linguistics . The IPA Japanese dictation toolkit project continued the work until 2000 when it was adopted by the Continuous Speech Recognition Consortium (CSRC) and from 2003 by the Interactive Speech Technology Consortium (ISTC). Initially only the Japanese language was supported, but thanks to the modular structure of the software, English language models were soon developed as well.

From version 3.4 onwards, a grammar module based on a finite automaton called Julian was integrated into the system, which is designed for the development of simpler voice control and dialogue systems. An embedded version for weaker processors has also been available since 2006 .

Related projects

Simon

The graphical front end Simon is currently in beta. It has been specially optimized for use by people with physical disabilities. The development of Simon is driven by the Simon Listens Association , which was founded in Austria in 2007 .

The front end Simon uses its own language model and can therefore also be operated using the German language. The model used by Simon is speaker-dependent and must be trained by the respective user in order to achieve a high recognition rate. After the end of the funding period, Peter Grasch continued to develop Simon up to version 0.4 and transferred it to the Linux desktop KDE as open source speech recognition software.

VoxForge

The VoxForge project promotes the development of free language models for Julius and similar platforms (such as CMU Sphinx ).

License terms

Julius is distributed under its own license, which grants the usual liberal rights of use for free software (free distribution and modification, availability of the source code). It has a copyleft clause; Further developments of the program may only be published under the same, unmodified license. In contrast to the GPL , the license specifies that the name of the program must be mentioned when new developments are made with its help.

Web links

Individual evidence

  1. ↑ Release 4.5 . January 2, 2019 (accessed January 3, 2019).
  2. ↑ a b License from Julius
  3. ^ Embedded Julius: Continuous Speech Recognition Software for Microprocessor , IEEE.org
  4. ↑ Simon 0.2 goes into the final spurt , Pro-Linux.de, January 13, 2009
  5. ↑ KDE Application Simon (2015) URL: https://kde.org/applications/unmaintained/org.kde.simon (last accessed 2019/12/23)