Application of russian language phonemics to generate macedonian speech recognition model using sphinx

Abstract

Speech-recognition provides possibilities for improved user experience and new level of features in various applications. Although there are widely available open-source and proprietary systems for speech recognition and synthesis for the more widely used languages, there are no available and robust enough systems for the Macedonian language. Building a speech recognition system requires a lot of high-quality recordings, which is expensive operation. To overcome this issue we propose applying some background knowledge and using existing speech synthesis engines to train speech recognition system. Since the Macedonian language belongs to the group of Slavic languages, there are many similarities between them. In this paper we apply this fact to generate a speech synthesis module for the Macedonian language based on the Russian language model. Furthermore, we use the speech synthesis module to build a speech recognition module using the CMUSphinx Toolkit. Finally the results are presented and they confirm that a system with substantial quality can be built without the need of manual recordings in the specific language.