The speech signal is a complex acoustic event. Beside conveying linguistic content, it also provides information about the speaker’s identity, personality, and emotions. This information is closely related to the quality of the speech sound. Speech quality itself is diversified since it gives rise to a variety of different perceptions among listeners.

Speaker-specific aspects of speech sound are mainly attributable to anatomical differences of speech organs, and a number of investigations have been made on the acoustic  correlates of speaker-identifying characteristics (1,2,3), as they provide data for extraction of the personality and are useful for speaker-normalization in a speaker-independent speech recognition scheme. Other aspects of research relating to speech quality are the analysis and perceptual nature of singing voices (4, 5) analysis of naturalness for speech synthesis (6,7,8), and the pathological study of abnormal voices (9). However, studies on the quality of normal voices (10,11) are relatively rare, because it is rather difficult to find direct applications to such speech technologies as synthesis and recognition.

The present study is an investigation designed to find acoustic parameters that contribute to the perception of voice-quality. Sustained tokens of the vowel [a] were used as the speech material, and psychophysical measurements were made at first based on the similarity judgment of several voice-qualities. Then, the speech samples were analyzed to extract such acoustic parameters as fundamental frequency  and formant frequencies using the inverse-filter method based on LPC analysis, and a  correlation analysis was subsequently  performed between these acoustic parameters  and the data of the similarity judgments.  Combinations of acoustic parameters are  important in investigating the relationship between these parameters and voice-qualities. Our aim, however, is to find the individual contributions of these parameters to the voice-qualities.

The quality of a sustained vowel is considered to be free from the speaker’s way of pronouncing a sentence and his/her dialect and linguistic background; it is mostly attributable to the anatomical structures of the vocal organs. In this study, therefore there was no need to consider linguistic parameters other than the acoustic ones that might reflect the speaker’s vocal tract configuration.