Speech quality assessment in low-bit-rate speech coding taking into consideration quality variation among speakers
Hiromi Nagabuchi · The Journal of the Acoustical Society of America · 1988
A method for assessing the quality of low-bit-rate speech coding is proposed. This method considers the speaker dependency of coded speech quality, which has not been done before. First, it is shown that coded speech quality varies with speakers more in low-bit-rate coding than in PCM. Next, the power of the linear prediction error of the signal of speech weighted by speech power, Gopt, is introduced as a measure for estimating quality variation among speakers. It is shown that Gopt, can be used as an effective index for selecting speakers in coded speech quality evaluation. Finally, a speech quality assessment method is proposed in which (1) test speakers are selected to cover uniformly the whole range of the Gopt, distribution; and (2) the assessment results are expressed in terms of not only the average but also the standard deviation of measured values for each speaker weighted according to the Gopt distribution characteristics. This method can assess the performance of low-bit-rate speech coding more accurately due to the increased reproducibility of assesed values when applied to different speakers.