Automatic detection of prominent words in Russian speech

An experimental research with a goal to automatically detect prominent words in Russian speech is presented in this paper. The proposed automatic prominent word detection system could be further used as a module of an automatic speech recognition system or as a tool to highlight prominent words within a speech corpus for unit selection text-to-speech synthesis. The detection procedure is based on the use of prosodic features such as speech signal intensity, fundamental frequency and speech segment duration. A large corpus of Russian speech of over 200 000 running words was used to evaluate the proposed prosodic features and statistical method of speech data processing. The proposed system is speaker-independent and achieves an efficiency of 84.2 %.