Ukrainian parliament speech recognition
A short history of speech recognition in Kiev is presented: from pioneer mathematical models based on dynamic programming algorithm to DSP based speech dialog devices.
An Extra large vocabulary speech recognition algorithm was developed and it is described in the presentation.
Ukrainain parliament speech (UPS) corpus was collected and annotated in the project. Development and evaluation of ASR methods on the UPS corpus is presented.
Precious transcriptions of various types of speakers was investigated to take in a consideration different speakers' pronunciations. Online speaker adaptation for multispeaker environments was developed.
Implementation of key words spotting technique to search of speech segments from the large speech corpora was developed and implemented. Demo application for Ukrainian-Russian speech will be presented in the talk.
Paralinguistic and nonverbal cues were analysed to improve spontaneous speech recognition. Applying speech document retrieval technique extends a total vocabulary up to 300K. Composed acoustic and pronunciation models were applied for multilingual ASR (Russian-Ukrainian in our case).
International Research/Training Center for
Information Technologies and Systems
Montag, 4. März 2013, 16 Uhr c.t.
Otto-von-Guericke-Universität Magdeburg G26.1-010 (Videoübertragung zur Universität Ulm, O28, Raum 1002)