The paper is devoted to improving the functionality of the speech intelligibility assessment solution. A previously implemented approach based on the Google Cloud Speech-to-Text solution has shown controversial quality indicators due to its attachment to the global news context. The paper proposes and evaluates the use of the Kaldi system for recognition. This solution not only made it possible to assess the quality of speech without using the Internet, but also to improve the quality of recognition in isolated phrases by excluding the global news context.