About Julius

• Julius is a high performance continuous speech recognition software based on word N-grams. It is able to perform recognition at the sentence level with a vocabulary in the tens of thousands of words.
• Julius realizes high-speed speech recognition on a typical desktop PC. It performs at near real time and has a recognition rate of above 90% for a 20,000 word vocabulary dictation task.
• The best feature of the Julius system is that it is multipurpose. By recombining the pronunciation dictionary, language and acoustic models one is able to build various task specific systems. The Julius code is also open source so one should be able to recompile the system for other platforms or alter the code for one's specific needs.
• Platforms currently supported include Linux, Solaris and other versions of Unix, and Windows. There are two Windows versions a Microsoft SAPI 5.0 compatible version and a Windows DLL version.
• This documentation relates to the Unix version of Julius. For documentation on the Windows versions see the Julius for SAPI README (CSRC CD-ROM) (Japanese), or the Julius for SAPI homepage (Kyoto University) (Japanese).

Contacts/Links

Contacts/Links

• Home Page: http://winnie.kuis.kyoto-u.ac.jp/pub/julius/ (Japanese)
• E-mail: julius@kuis.kyoto-u.ac.jp

Developers

• Original System/Unix version: Akinobu Lee (ri@is.aist-nara.ac.jp)
• Windows Microsoft SAPI version: Takashi Sumiyoshi
• Windows (DLL version): Hideki Banno