A Methodology for Improving PESQ accuracy for Chinese Speech

Abstract

Unlike English and most other European languages, Mandarin Chinese has two unique characteristics, the consonant-vowel-consonant (CVC) phonetic structure and use of tones, which may affect its intelligibility after processing by sound processing systems. Due to this, the perceptual evaluation of speech quality (PESQ) objective speech quality measurement system, which has been proven effective in measuring the speech quality of sound processing systems processing English or some other languages, may not accurately measure speech quality of systems processing Chinese speech. An evaluation was thus performed with PESQ to investigate whether intelligibility related problems that arise from the two characteristics are being considered in the computation of speech quality. Our evaluation reveals that PESQ indeed does not consider them through low correlation between subjective intelligibility and PESQ scores. A method known as consonant amplification was proposed to improve correlation results for Chinese speech, and this method is evaluated with PESQ.