Research On The Relationship Between Acoustic Distance And Perceptual Distance Of Chinese Initials

Since the acoustic space and perceptual space are not equivalent to each other, it’svery important to investigate the relationship between the acoustic space andperceptual space of Chinese, Tibetan, and Uighur. It was found out that the perceptionof Chinese initials is more difficult than that of Chinese finals. Many works have beendone in the methods of the perception measurements for speech sounds. However,most of them are subjective measurement alone for perception aspects. In this paper,the monotonic relation between acoustic features and perceptual measurement is builtand a new method for the objective perceptual measurement is given.Firstly, different types of acoustic features including frequency-domain andtime-domain features are investigated. A couple of experiments with respect toacoustic space and perceptual space are conducted. In order to single out whichdistance metrics are suitable for measuring the acoustic distance and which acousticfeatures or their combinations are most consistent with the chosen distance metrics, aseries of clustering analyses are conducted by using Mel Frequency CepstralCoefficient and Perceptual Linear Prediction. Based on the chosen acoustic featuresand distance metrics, the acoustic distance between different types of initials isdefined. Then, the perceptual distance is defined and a speech audiometry isconducted to obtain the confusion matrix, which can be transformed into theperceptual matrix by using the definition of the perceptual distance. Finally, thechosen acoustic features and distance metrics are validated by using a nonparametricmethod called Spearman’s Rank Correlation Coefficient. The most consistenttime-domain features are singled out by using the same procedures. Based on thechosen frequency-domain and time-domain features, a new formula is proposed tomeasure the acoustic distance objectively, which has a monotonic relation with theperceptual distance of different types of Chinese initials.The method proposed in this paper can be not only the foundation of the purposeto describe the acoustic space and do the research on the perception of speech soundson syllable level, but also used to automatically evaluate the difficulty of the wordlists in speech audiometry.