A Speech Intelligibility Estimation Method Based on Hidden Markov Model

Abstract

This paper proposes a speech intelligibility estimation method based on hidden Markov model (HMM) that is widely used for speech recognition. The HMM-based method is a kind of non-intrusive speech quality measurement, which means it operates without a reference speech signal. The log-likelihood score of HMM is converted to a normalized intelligibility score. We estimate the speech intelligibility of standard digital speech coders. The experimental results show that the proposed HMM-based method gives improved performance than the conventional non-intrusive speech intelligibility evaluation tool.