Multigranular scale speech recognition: tehnological and cognitive view