Thai Tones Research Articles

Effects of native language and training on speaker normalization in lexical tone perception

Tone languages, such as Thai, use pitch differences to distinguish lexical meaning. In this study, we investigated the effects of native language and training on speaker normalization in perception of lexical tones in Thai. Three tokens each of the Thai high, mid, and low tones were generated to simulate three female speakers with three relatively different F0 pitch ranges (high, mid, and low). The stimuli were presented to native Mandarin Chinese (NC), American English (NE), and Thai (NT) for categorization before and after a two‐day perceptual training. Results obtained thus far from 32 NC, 30 NE, and 11 NT suggested that, overall: (a) NC outperformed NE both before and after training, (b) NT numerically outperformed NC before training, but not after training, and (c) performance of all three groups of participants improved after training. Interestingly, it was also found that participants’ performance varied from tone to tone, depending on the tonal environment in which it was presented for discrimination. These results suggest that knowledge of a tonal language and perceptual training facilitates the speaker normalization process in lexical tone perception, and that tonal contexts affect discrimination of tones.

Consonant-Tone Interaction in Thai: An OT Analysis

Thai tones have traditionally been characterized in terms of a correlation between tones and syllable type (Gandour 1974, Tumtavitikul 1993, Intrasri 2001, Morin and Zsiga 2006). However, I show that a more interesting set of observations can be made on the distribution of Thai tones in which they can be explained as a consequence of consonant-tone interaction within an OT framework. The observations are that unaspirated stops are not compatible with a high tone vowel in the nucleus. An obstruent coda has the effect of shortening the vowel in the nucleus, something that is testable via phonetic experimentation. This shortening effect places a burden on either phonetic perception or production, so that the five-way contrast is neutralized to a two-way contrast. The shortening effect places an especially strong burden on contour tones, so these tones are preferably excluded from the two-way contrast. The best two-way contrast among the three level tones is a simple high-low contrast, which is more readily perceivable than a low-mid or high-mid contrast. I also show that unaspirated obstruents prefer to be adjacent to a non-high tone. Unaspirated coda are allowed to be adjacent to a preceding high tone in a short vowel, but a long vowel presents an opportunity to insert a low tone on the second tone bearing unit, thus producing a two-way contrast between low and falling tone instead. This insertion of a low tone is done in order to avoid a situation where a high tone vowel would be adjacent to the unaspirated consonant. A high-tone vowel-unaspirated coda sequence would result otherwise. It is argued that the attested patterns of consonant-tone interaction in Thai are captured by conjoining two markedness constraints in addition to the simple markedness constraints.

The Lexical and Post-Lexical Phonology of Thai Tones*

At first glance, the five-tone system of Thai looks quite simple. However, a detailed examination of the phonological distributions of segments and tones, combined with a careful analysis of the phonetic realizations of duration and pitch in both citation forms and connected speech, lead to the conclusion that the system is in fact complex and interesting. Based on data from an acoustic experiment, we claim that the tone bearing unit in Thai is the mora, and that previously unexplained restrictions on the distributions of tones in syllables closed by obstruents are the result of a relationship between the glottal feature and low tones. We describe and explain unexpected differences in the realization of tones in different phrasal positions in connected speech, and show that there are non-neutralizing contour tone simplifications that take place non-finally at the post-lexical level. Our analysis combines descriptive phonetics and phonology with both representational and constraint-based explanations (incorporating positional faithfulness and stratal Optimality Theory) to provide a unified account of the Thai tonal system. Our study supports a view of the grammar in which phonetics and phonology are separate, yet intricately related.

Synthesis of Polysyllabic Sequences of Thai Tones Using a Generative Model of Fundamental Frequency Contours

In this paper, the distinctive tones of Thai in running speech are studied. We present rules to synthesize F0 contours of Thai tones in running speech by using the generative model of F0 contours. Along with our method, the pitch contours of Thai polysyllabic words, both disyllabic and trisyllabic words, were analyzed. The coarticulation effect of Thai tones in running speech were found. Based on the analysis of the polysyllabic words using this model, rules are derived and applied to synthesize Thai polysyllabic tone sequences. We performed listening tests to evaluate intelligibility of the rules for Thai tones generation. The average intelligibility scores became 98.8%, and 96.6% for disyllabic and trisyllabic words, respectively. From these result, the rule of the tones' generation was shown to be effective. Furthermore, we constructed the connecting rules to synthesize suprasegmental F0 contours using the trisyllable training rules' parameters. The parameters of the first, the third, and the second syllables were selected and assigned to the initial, the ending, and the remaining syllables in a sentence, respectively. Even such a simple rule, the synthesized phrases/senetences were completely identified in listening tests. The MOSs (Mean Opinion Score) was 3.50 while the original and analysis/synthesis samples were 4.82 and 3.59, respectively.

Training English and Chinese Listeners to Perceive Thai Tones: A Preliminary Report

The ability of native English (NE) and native Chinese (NC) speakers to identify and discriminate the mid‐ versus the low‐tone contrast in Thai was investigated before and after auditory training. The variables under investigation were first language background and the interstimulus interval (ISI) of the presentation (500 ms vs. 1500 ms). The NC group outperformed the NE group in its ability to discriminate the two Thai tones under the ISI 500 ms condition before training and under both ISI conditions after training. A significant improvement in identification from the pretest to the posttest was observed in the NC group under both ISI conditions, but not in the NE group. These results suggest that prior experience with the tone system in one tone language may be transferable to the perception of tone in another language.

SUPPORT VECTOR MACHINES FOR THAI PHONEME RECOGNITION

The Support Vector Machine (SVM) has recently been introduced as a new pattern classification technique. It learns the boundary regions between samples belonging to two classes by mapping the input samples into a high dimensional space, and seeking a separating hyperplane in this space. This paper describes an application of SVMs to two phoneme recognition problems: 5 Thai tones, and 12 Thai vowels spoken in isolation. The best results on tone recognition are 96.09% and 90.57% for the inside test and outside test, respectively, and on vowel recognition are 95.51% and 87.08% for the inside test and outside test, respectively.

Support Vector Machines for Thai Phoneme Recognition

The Support Vector Machine (SVM) has recently been introduced as a new pattern classification technique. It learns the boundary regions between samples belonging to two classes by mapping the input samples into a high dimensional space, and seeking a separating hyperplane in this space. This paper describes an application of SVMs to two phoneme recognition problems: 5 Thai tones, and 12 Thai vowels spoken in isolation. The best results on tone recognition are 96.09% and 90.57% for the inside test and outside test, respectively, and on vowel recognition are 95.51% and 87.08% for the inside test and outside test, respectively.

Tone Recognition of Continuous Thai Speech Under Tonal Assimilation and Declination Effects Using Half-Tone Model

This paper presents a method for continuous Thai tone recognition. One of the main problems in tone recognition is that several interacting factors affect F0 realization of tones. In this paper, we focus on the tonal assimilation and declination effects. These effects are compensated by the tone information of neighboring syllables, the F0 downdrift and the context-dependent tone model. However, the context-dependent tone model is too large and its training time is very long. To overcome these problems, we propose a novel model called the half-tone model. The experiments, which compare all tone features and all tone models, were simulated by feedforward neural networks. The results show that the proposed tone features increase the recognition rates and the half-tone model outperforms conventional tone models, i.e. context-independent and context-dependent tone models, in terms of recognition rate and speed. The best results are 94.77% and 93.82% for the inside test and outside test, respectively.

TONE RECOGNITION OF CONTINUOUS THAI SPEECH UNDER TONAL ASSIMILATION AND DECLINATION EFFECTS USING HALF-TONE MODEL

This paper presents a method for continuous Thai tone recognition. One of the main problems in tone recognition is that several interacting factors affect F0realization of tones. In this paper, we focus on the tonal assimilation and declination effects. These effects are compensated by the tone information of neighboring syllables, the F0downdrift and the context-dependent tone model. However, the context-dependent tone model is too large and its training time is very long. To overcome these problems, we propose a novel model called the half-tone model. The experiments, which compare all tone features and all tone models, were simulated by feedforward neural networks. The results show that the proposed tone features increase the recognition rates and the half-tone model outperforms conventional tone models, i.e. context-independent and context-dependent tone models, in terms of recognition rate and speed. The best results are 94.77% and 93.82% for the inside test and outside test, respectively.

Discrimination of Thai tones by naive and experienced native English speakers

This study investigated the ability to discriminate between the low and the mid tones of Thai by two groups of native English speakers: naive and experienced. The naive group was comprised of eight native American English speakers who had no previous exposure to Thai or any tonal language. The eight native English speakers in the experienced group had been learning and had lived in Thailand for a varying amount of time. Eight native speakers of Thai participated as control subjects. Stimuli were eight minimal pairs. Three variables were manipulated: (1) experience with the target language: naive vs experienced; (2) syllable types: closed vs open; and (3) levels of processing: phonetic vs phonological implemented by using two different ISI (500 ms and 1500 ms). It was found that experience with the target language resulted in improved discrimination of the target language’s tonal contrast. Moreover, all three groups of subjects performed better with closed syllables than open syllables. Contrary to previous research, the effect of ISI was found only for the most difficult contrasts and only for the experienced group. An explanation for the findings and implication for futher research will be offered and discussed.

Effects of Speaking Rate on Thai Tones

Effects of speaking rate on the production of lexical tones (mid, low, falling, high, rising) were investigated in Thai. Stimuli consisted of bisyllabic adverbials (first syllable unstressed, last syllable stressed) elicited in a fixed syntactic and prosodic environment. For unstressed and stressed syllables separately, F₀- and time-normalized, fitted, third-order polynomial curves were used to compare height and slope characteristics of the tonal contours at each of 11 measurement locations between fast and slow speaking rates. Results indicated that speaking rate effects on F₀ contours of unstressed syllables are more extensive, both in terms of height and slope, than those of stressed syllables. In particular, the height of F₀ contours in unstressed syllables was generally higher in the fast speaking rate when compared to the slow. Analysis of the preceding carrier syllable revealed that changes in height of F₀ contours of unstressed syllables may be due primarily to perseverative effects of tonal coarticulation rather than to stress itself. The slope of F₀ contours in unstressed syllables varied depending on range of F₀ movement. Thai tones with substantial F₀ movement (falling, high, rising) exhibited overall flatter slopes at the fast speaking rate; those tones with lesser F₀ movement (mild, low) displayed steeper slopes. Despite extensive changes in height and shape, the five-way tonal contrast appears to be maintained in unstressed syllables at a fast speaking rate albeit in a different tonal space.

On the mechanism of lowering the voice fundamental frequency in the production of tones of Standard Chinese

While it is well known that the cricothyroid (CT) muscle is mainly responsible for lowering the voice fundamental frequency (F0) in many languages, the mechanism for F0 lowering in languages such as Standard Chinese (SC) has not been elucidated. Although several studies have shown that the sternohyoid (SH) muscle activity is strongly correlated with F0 lowering in SC, the mechanism itself is not clear since SH is not directly attached to the thyroid cartilage, whose movement is essential in changing the length and tension of the vocal chord. On the basis of an earlier finding on the production of Thai tones [D. Erickson, Annual Bulletin No. 27, RILP, Univ. Tokyo, pp. 135–149 (1993)], the present author has suggested the active role of the thyrohyoid (TH) muscle in F0 lowering in languages such as SC and Swedish [H. Fujisaki, Proc. XXIII World Congress of the International Association of Logopedics and Phoniatrics, pp. 156–159 (1995)]. The present study shows the detailed mechanism for F0 lowering involving both TH and SH in the production of the second, third, and fourth tones of SC based on new electromyographic observations.

Classification of Thai tone sequences in syllable-segmented speech using the analysis-by-synthesis method

Tone classification is important for Thai speech recognition because tone affects the lexical identification of words. An analysis-by-synthesis algorithm for classifying Thai tones in syllable-segmented speech is presented that uses an extension to Fujisaki's (1983) model for tone languages that incorporates tonal assimilation and declination. The classifier correctly identifies all of the tones in 89.1% of the test utterances.

Structural and functional aspects of tone split in Thai

Comparative and diachronic work throughout this century has laid a very strong foundation for the hypothesis that Thai owes its great number of tones to processes which created tones and splitted existing tones at an earlier stage of the language, and the further hypothesis that those processes had a direct connection with consonant mutations which involved segmental loss of contrast. Since the sixties, studies of Thai tones have emphasize the phonetic aspects of tonogenesis or tone split (cf. the key-note paper of Abramson and Erickson 1991). It is argued in the present paper that although the hypotheses stated above are not at issue, the diachronic scenario is still poorly understood, both with respect to the underlying phonetic mechanisms and with respect to the structural and functional aspects of the sound changes. It is a stumbling-stone for the formulation of phonetic explanations of tone split that there are quite different types of consonants involved, and with quite different diachronic behaviour. The paper suggests that higher-level (cognitive-phonological) categorization of syllable types may have played a central role in the implementation of tone split. Another issue is the striking variation in tone systems across the whole gamut of Thai dialects. The paper takes issue with the general assumption that early Thai developed a grossly over-complex tone system which was subsequently drastically reduced, and that the dialectal variation essentially reflects alternative simplifications. It is argued in this paper that the tone splits which we can observe in the modern dialects, are individual and functionally optimal solutions to the preservation of contrast between syllable types. It is further argued that there is no compelling reason to project tone split back to a period which was prior to the breaking-up of Thai into its present main dialects. Tone split may have spread geographically across dialects at a later date, each dialect responding according to its specific structural properties.

Interaction between Tone and Intonation in Thai after Unilateral Brain Damage

Intonational characteristics of Thai sentences were used to evaluate fundamental frequency (F0) control in brain-damaged patients with unilateral left and right hemisphere lesions. Subjects (n= 41) included 9 young and 10 old normal adults, 12 right hemisphere patients, and 10 left hemisphere aphasic patients (7 fluent and 3 nonfluent). Sentences were comprised of six words, three of which were keywords occurring in sentence-initial, -medial, and -final positions. All 125 possible sequences of three of the five Thai tones (mid, low, falling, high, rising) were superimposed on monosyllabic keywords. Utterances were produced at a conversational speaking rate. AverageF0of keywords was analyzed as a function of sentence position, tone, and group. For both normal and brain-damaged speakers, results indicated that tones in sentence-final position were significantly lower inF0than in either sentence-initial or -medial position; falling and high tones were significantly higher inF0than mid, low, and rising tones. Findings are discussed in relation to issues pertaining to hemispheric specialization and the nature ofF0deficits in nonfluent and fluent aphasic patients.

Tonal Coarticulation in Thai after Unilateral Brain Damage

The magnitude and temporal extent of anticipatory and perseverative tonal coarticulation was investigated in Thai-speaking normal and brain-damaged adults. A total of 47 speakers (10 young normal, 10 old normal, 13 nonaphasic right-brain-damaged patients, 14 left-brain-damaged aphasic patients, 9 fluent, 5 nonfluent) produced all 25 possible sequences of two tones from the five lexical tones of Thai embedded in a carrier sentence.F0contours were analyzed in terms of height and slope at 10% intervals throughout the duration of the two syllables. Acoustic analysis revealed that anticipatory and perseverative tonal coarticulation of tones was markedly reduced in left fluent aphasics, totally absent in left nonfluent aphasics, but reasonably intact in right hemisphere patients. Findings are interpreted to highlight the nature of speech disturbances in nonfluent and fluent aphasia, hemispheric specialization for tone, and tonal coarticulation in Thai.

A case study of the effect of adventitious deafness on perception and production of Thai tones

This study investigated the perception and production of Thai tones by an adult, native speaker of Thai with a severe to-profound sensorineural hearing impairment. Five Thai words minimally distinguished by tone were presented for identification in both natural speech and synthetic speech versions. The same five words were produced by the hearing-impaired speaker. Results indicated that both tone perception and production were impaired, but that the magnitude of the deficit was greater in perception.

Tonal coarticulation in Thai

In tones, coarticulatory effects occur in much the same way as they do between consonants and vowels. Earlier studies on tonal coarticulation leave open the question as to whether the primary direction of coarticulatory effects is the same across tone languages, and whether the effect involves primarily a change in tone height or a change in slope as well. The present study investigated the magnitude and extent of tonal coarticulation in Thai. Ten speakers produced all 25 possible sequences of two tones from the five tones of Thai embedded in a carrier sentence. Tonal contours were analyzed in terms of height and slope of the fundamental frequency. Results from statistical analysis indicated that the coarticulatory effects were not symmetric. Thai tones were more influenced by carry-over than by anticipatory coarticulation. Carry-over coarticulation affected a greater number of Thai tones and extended farther into adjacent tones. Coarticulation affected primarily height; slope was relatively unaffected. Magnitude of the effects was fairly uniform regardless ofdirection of coarticulation.

Anticipatory Tonal Coarticulation in Thai Noun Compounds after Unilateral Brain Damage

The time course and extent of anticipatory coarticulation between tones was investigated in normal and brain-damaged Thai-speaking subjects. Subjects were classified into five groups including 11 young normal, 9 old normal, 12 right hemisphere, 9 left hemisphere fluent, and 6 left hemisphere nonfluent. Stimuli consisted of five bisyllabic noun compounds with a falling tone on the initial syllable and each of the five Thai tones (mid, low, falling, high, rising) on the final syllable. Height and slope of F 0 was measured at 10% intervals throughout the duration of the initial syllable. Results indicated anticipatory effects on both height and slope of the falling tone. Height effects extended throughout from the beginning. The falling tone was generally higher when occurring before the low/rising tones than when occurring before the mid/falling/high tones. Slope effects were restricted to the terminal portion. The falling tone before the low/rising tones exhibited a steeper slope than before falling/high tones. In magnitude of effect, patients with left and right hemisphere lesions were statistically indistinguishable from those of normal subjects. No differences were noted in coarticulatory patterns as a function of aphasia type. All brain-damaged speakers were more variable in F 0 production than normals. Findings are interpreted to highlight properties of nonfluent aphasic speech and neurological bases of speech production.

Lexical tones in Thai after unilateral brain damage

An acoustic perceptual investigation of the five lexical tones of Thai was conducted to evaluate the nature of tonal disruption in patients with unilateral lesions in the left and right hemisphere. Subjects ( n = 48) included 10 young normal adults, 10 old normal adults, 11 right hemisphere nonaphasics, 9 left hemisphere fluent aphasics, and 8 left hemisphere nonfluent aphasics. The five Thai tones (mid, low, falling, high, rising) were produced in isolated monosyllables, presented for tonal identification judgments, and measured for fundamental frequency (F o) and duration. Results of an analysis of variance indicated that left hemisphere nonfluent speakers signaled tonal contrasts at a lower level of proficiency. The extent of their impairment varied depending on severity level of aphasia. When compared to normal speakers, tonal identification for less severe nonfluent aphasics differed more in degree than in kind, and for more severe nonfluent aphasics differed both in kind and in degree. Acoustic analysis revealed that with the exception of one left nonfluent, average F o contours were comparable in shape across speaker groups. Variability in F o production, however, was greater in left nonfluent speakers than in any of the other four groups of speakers. Issues are discussed regarding the extent and nature of tonal disruption in aphasia and hemispheric specialization for tone production.