| |
 |
 |
Since 1978
Published in Sarov (Arzamas-16), Nizhegorodskaya oblast |
RUSSIAN FEDERAL
NUCLEAR CENTER -
ALL-RUSSIAN RESEARCH INSTITUTE
OF EXPERIMENTAL PHYSICS |
|
 Русский |  English
|
SDT - SOFTWARE PACKAGE FOR CONTINUOUS SPEECH RECOGNITION
V.A. Barannikov, Alexander A. Kibkalo
VANT. Ser.: Mat. Mod. Fiz. Proc 2003. Вып.3. С. 3-7.
SDT (Speech Developer Toolkit), the system for large vocabulary continuous speech recognition (LVCSR), was developed in VNIIEF- STL project. The package was used to build US English and Mandarin speech recognition systems. SDT functionality includes feature extraction, acoustic models building and adaptation, fast decoding with finite and stochastic grammar, speaker recognition, decoding results scoring. The last version of SDT includes some features to support LVCSR for the Russian language. The decoding results for Russian tasks with closed and open dictionaries are presented.
| DEVELOPMENT OF RUSSIAN SPEECH RECOGNITION SYSTEM
Alexander A. Kibkalo, M.M. Lotkov, I.G. Rogozhkin, A.A. Turovetz
VANT. Ser.: Mat. Mod. Fiz. Proc 2003. Вып.3. С. 8-20.
High-quality systems of continuous speech recognition for some languages are currently available. However the progress in this field for Russian and other Slavonic languages is rather moderate. The main reasons are the lack of high-quality Russian language resources and peculiarities of Russian language (large number of wordforms and arbitrary order of words in a sentence). The paper presents the latest results of the efforts to create a continuous speech recognition system with a large vocabulary, namely, recording of Russian audio packages and construction of acoustic models for a system prototype, creation of a large Russian phonetic vocabulary and development of a Russian speech decoder model. These activities are based on the package of STD 4.0 program developed by VNIIEF-STL firm.
|
[ Back ] |
|
|
| |
| |
 |
| |
|