Azerbaijan, Baku, July 4 / Trend G. Veliyev /
The first stage of operations within project of computer system of speech identification in Azerbaijani was completed, Dilmanc project head Abulfat Fatullayev told Trend on July 3.
System is capable to identify 130,000 words and expressions in Azerbaijani, Fatullayev said. Acoustic and Language modules are used in the system. Language module is built in corpus consisting of 150 million expressions but Acoustic module on 50 hour-speech corpus. Voices of 1,500 speakers were used in creation of the system.
The system is convenient for introducing the text via microphone. Language filter is tuned in accordance with pronunciation. It allows turning speech in text in any Windows edition.
Presently, speed of big sentences change into E-variant hits 15-16 seconds.
This index will improve according to improvement of the system in future.
Initial version of the system will be accessible at http://www.dilmanc.az by late 2009.
Do you have any feedback? Contact our journalist at [email protected]