PHONETIC AND ALLOPHONIC TEXT PROCESSING IN BELARUSIAN AND RUSSIAN SPEECH SYNTHESIZER FOR MOBILE PLATFORMS

Abstract

The article describes methods of «grapheme – phoneme» and «phoneme – allophone» conversions for Belarusian and Russian speech synthesis. For speech synthesizers on mobile platforms, the rule-based method has been selected. The article describes text processing algorithms at the input phase and rules for «grapheme – phoneme» and «phoneme – allophone» conversions, providing allophone chains for an acoustic processor. The developed algorithms were evaluated with respect to the reference-class stationary Multiphone text-to-speech synthesis system. The test data demonstrates a high accuracy level: over 80% for word processing, over 98% for phoneme processing and over 96% for allophone processing (for Belarusian and Russian languages). The subsequent work will be devoted to the improvement of practical realization of the developed algorithms.