ICPhS XVII Regular Session Hong Kong, 17-21 August 2011 THE PHONATION FACTOR IN THE CATEGORICAL PERCEPTION OF MANDARIN TONES Ruo-Xiao Yang The Chinese University of Hong Kong, Hong Kong [email protected] ABSTRACT Tone is a critical element for tone languages.. Generally, tone is produced by the vibration of This study presents a new way of observing the vocal folds, which is often measured as categorical nature of Mandarin tone perception by fundamental frequency (F0) acoustically and considering the phonation factors in the perceived as pitch. When tone is described or experiment design. In order to demonstrate the perceived, the F0, sometimes as well as intensity reliability of experiment materials, a pitch and duration, is taken as its most critical physical synchronous overlap-add (PSOLA) method is used correlates. However, this standpoint is only based to produce the speech stimuli, and an inverse- on the physical or acoustic representations of tone filtering method is applied to assess the but overlook its articulator, the vocal folds. invariability of sound source of the stimuli. Results Vocal folds are the source of each sound, which provide evidence for the prediction that phonation connect the dynamics from lung beneath and the factors are adopted in perceiving tones by resonance from vocal tract above. The vibration of Mandarin speakers. Especially for the tone which vocal folds for speech production can be described has a special voice characteristic, such as the third not only as its vibrating velocity per second, i.e. tone in Mandarin, the perception difference among the fundamental frequency (F0), but also its continua with different voice qualities is much vibrating mode which is often referred as more obvious. Results also suggest that there is phonation types (or loosely, voice qualities). Since categorical perception for each pair of two tones in a speech sound can be divided into segmental Mandarin. These results lead to a discussion about (consonants and vowels) and super-segmental parts the necessity of a wider and more integrated theory under the backgrounds of modern phonology and model for describing tones in tone languages, in phonetics, the phonation part should be sorted to which the phonation factors should be considered. be a super-segmental feature. Keywords: phonation perception, voice quality, Related to surveys on the CP nature of tones, categorical perception, Mandarin tones the usual pervious hypothesis has suggested that F0 is the main or even the most crucial correlate 1. INTRODUCTION for perceiving tones. Perception experiments are Categorical perception (CP) is an important thus designed based on this viewpoint. However, if phenomenon in human cognition because people tone is observed from the “phonation” perspective, sort out surrounding things incessantly every day. F0 might not be the only correlate for Roughly, the differences among things can be distinguishing tones as different phonemic units. perceived with two modes: the “continuous Furthermore, more and more evidence show that perception” and the “categorical perception (CP)”. some tone languages (such as Hani language and CP have been formally thought to be peculiar to Jingpo language) indeed use different phonation speech and color of human perceivers, but it turns types as distinctive features rather than F0/pitch out to be much more general by later research on (Kong, [6]). Thus, it is important and necessary to human infants and animals [4, 8, 9]. CP has been discuss the CP nature of tones within a phonation investigated in different levels of language. In perspective. human speech, the general conclusion about Several previous studies have explored the vowels as continuous perception and consonants as phonation modes of Mandarin tones from a linguistic categorical perception has been widely accepted, viewpoint. Kong used EGG (Electroglottographic) whereas the categorical nature of tone in tone signal [6] and high-speed digital imaging [7] to languages is still controversial. investigate the voice of the four basic tones of Mandarin/standard Chinese in single syllables in 2204 ICPhS XVII Regular Session Hong Kong, 17-21 August 2011 isolated monosyllabic words. These research controlled and thus only [ta] syllable was used. suggested that F0 and phonation features all Previous studies suggest that [ta] is the best contribute to the perception of Mandarin tones. syllable structure for keeping the formant Keating and Esposito [5] also took some preliminary information when manipulating F0. measurements on F0 and other parameters related to For assessing the phonation information during phonation types of the four basic tones of Mandarin. perceiving Mandarin tones, 12 10-step continua They also pointed out some special phonation were synthesized as stimuli by manipulating F0 features in the low falling tone and the end of the through the “pitch synchronous overlap-add” falling tone of Mandarin, i.e., the creaky voice can be method (i.e., PSOLA) in Praat [1]. heard on the low falling tone and visible at the end of Table 1: Syllable with four Mandarin tones and F0 the falling tone. Moreover, they suggested a more parameters of four tones for synthesis. general conclusion that all “tones” (in all languages) Four Tones OT F0 (Hz) TTP F0 (Hz) TT F0 (Hz) may have some correlated variation in phonation, only most of them being subtle within the modal Tone1 (T1) 160 160 160 Tone2 (T2) 100 123 160 ranges and some owning phonological status, but it Tone3 (T3) 90 80 130 was probable that listeners in many tone languages Tone4 (T4) 160 130 80 use the phonation information in recognizing their tones especially out of context. Even though the Figure 1 (A-F): Diagrams of manipulating F0 in 12 relationships of phonation and tones have attracted continua: (A) T1T2/T2T1, (B) T1T3/TT1, (C) T1T4/T4T1, (D) T2T3/T3T2, (E) academic attention, no studies have been done to T2T4/T4T2, (F) T3T4/T4T3. discuss how the phonation information related to tones can be perceived. Moreover, no previous T1-->T2 & T2-->T1 T1-->T3 & T3-->T1 T1-->T4 & T4-->T1 170 170 170 160 160 160 research on CP of tones and perceiving phonation 150 150 150 140 140 140 130 130 130 120 120 120 F0(Hz) F0(Hz) 110 information has attempted to consider the phonation F0(Hz) 110 110 100 100 100 90 90 90 information contributing to perception of tones. We 80 80 80 OT TTP TT OT TTP TT OT TTP TT argue that it is important to look at the phonation Time Points Time Points Time Points factor and CP of tones concurrently in one (A) (B) (C) experiment design in order to achieve a better T2-->T3 & T3-->T2 T2-->T4 & T4-->T2 T3-->T4 & T4-->T3 understanding of CP and how to define the 170 170 170 160 160 160 150 150 150 140 140 140 phonological status of tones in tone languages. 130 130 130 120 120 120 F0(Hz) F0(Hz) F0(Hz) 110 110 110 100 100 100 Therefore, in this experiment, the phonation factor 90 90 90 80 80 80 has been considered concurrently with CP testing. OT TTP TT OT TTP TT OT TTP TT Time Points Time Points Time Points This design permits us to examine whether CP exists in arbitrary two tones of the four basic tones in (D) (E) (F) Mandarin by making continua as stimuli between Taylor [11] explained that TD-PSOLA is each pairs of two tones and to determine whether attempting to mimic the process of safely changing phonation information is used in differentiating tone the pitch without changing filter by separating the categories by synthesized stimuli. effects of each pulse. Thus, under the most ideal 2. EXPERIMENT MATERIALS AND condition, TD-PSOLA will keep the spectral METHODS envelope characteristics and therefore keep the original phonation/voice information from the 2.1. Materials sound source. Furthermore, Lemmetty [10] and The stimuli were manipulated based on four Upperman [12] also proposed that the speech signal Mandarin syllables with four basic tones out of resulted from PSOLA has the same spectrum as the context. To make the syllable more real and keep original signal but with a different F0. Esposito [3] better phonation information (which may be lost in also suggested that PSOLA changes the F0 of a context), the four original syllables are produced signal without changing other properties of the separately by a male speaker whose native voice and applied it to normalize F0 of stimuli for language is Beijing Mandarin. Moreover, in order testing the perception of phonation types. to keep as much sufficient information of original 2.2. Subjects syllables as possible and avoid losing information during manipulation, the syllable structure was 19 participants from Peking University in China 2205 ICPhS XVII Regular Session Hong Kong, 17-21 August 2011 (11 female and 8 male, aged 19-25 years) the tone with this voice quality; therefore, it could participated for a small amount of money in the be roughly concluded here that voice quality or experiment: an identification test followed by a phonation information indeed influenced the discrimination test. All of them were native fluent categorical perception of Mandarin tones. speakers of Mandarin / Putonghua with different Figure 2 (A-F): Identification curves for responses of Chinese dialect backgrounds. 19 subjects in 12 continua to show the differences between pairs of continua with different voice 2.3. Experiment procedures qualities: (A) T1T2 based on the voice of T1 (solid line) and T2 (dashed line); (B) T1T3 based on the The whole experiment was consisted of an voice of T1 (solid line) and T3 (dashed line); (C) identification task and a AX discrimination task. In T1T4 based on the voice of T1 (solid line) and T4 the identification task, subjects listened to stimuli (dashed line); (D) T2T3 based on the voice of T2 presented in isolation.