Chutamanee Onsuwan

A Rule-Based Method for Thai Elementary Discourse Unit Segmentation (TED-Seg)

Discovering discourse units in Thai, a language without word and sentence boundaries, is not a straightforward task due to its high part-of-speech (POS) ambiguity and serial verb constituents. This paper introduces definitions of Thai elementary discourse units (T-EDUs), grammar rules for T-EDU segmentation and a longest-matching-based chart parser. The T-EDU definitions are used for constructing a set of context free grammar (CFG) rules. As a result, 446 CFG rules are constructed from 1,340 T-EDUs, extracted from the NE- and POS-tagged corpus, Thai-NEST. These T-EDUs are evaluated with two linguists and the kappa score is 0.68. Separately, a two-level evaluation is applied, one is done in an arranged situation where a text is pre-chunked while the other is performed in a normal situation where the original running text is used for test. By specifying one grammar rule per one T-EDU instance, it is possible to make the perfect recall (100%) in a close environment when the testing corpus and the training corpus are the same, but the recall of approximately 36.16% and 31.69% are obtained for the chunked and the running texts, respectively. For an open test with 3-fold cross validation, the recall is around 67% while the precision is only 25-28%. To improve the precision score, two alternative strategies are applied, left-to-right longest matching (L2R-LM) and maximal longest matching (M-LM). The results show that in the L2R-LM and M-LM can improve the precision to 93.97% and 94.03% for the running text in the close test. However, the recall drops slightly to 94.18% and 92.91%. For the running text in the open test, the f-score improves to 57.70% and 54.14% for the L2R-LM and M-LM.

An EDU-Based Approach for Thai Multi-Document Summarization and Its Application

Due to lack of a word/phrase/sentence boundary, summarization of Thai multiple documents has several challenges in unit segmentation, unit selection, duplication elimination, and evaluation dataset construction. In this article, we introduce Thai Elementary Discourse Units (TEDUs) and their derivatives, called Combined TEDUs (CTEDUs), and then present our three-stage method of Thai multi-document summarization, that is, unit segmentation, unit-graph formulation, and unit selection and summary generation. To examine performance of our proposed method, a number of experiments are conducted using 50 sets of Thai news articles with their manually constructed reference summaries. Based on measures of ROUGE-1, ROUGE-2, and ROUGE-SU4, the experimental results show that: (1) the TEDU-based summarization outperforms paragraph-based summarization; (2) our proposed graph-based TEDU weighting with importance-based selection achieves the best performance; and (3) unit duplication consideration and weight recalculation help improve summary quality.

Perception of Thai distinctive vowel length in noise

A forced choice identification perception experiment using 150 monosyllabic rhyming-word stimulus pairs (with identical consonants and tone) in four conditions of white Gaussian noise was conducted to explore vowel confusions in Thai, a language with nine monophthongs and length (short-long) contrast for all vowels (e.g., /i/-/i:/ and /o/-/o:/). Each stimulus containing speech and noise portions is equal in length. Perceptual results of 18 vowels from 36 Thai listeners at a noise level (SNR) of -24 dB, where the percent intelligibility is the most interpretable, showed that stimuli with short vowels are more accurately perceived than those with long vowels (93.46 vs. 85.64%) with /o:/ and /e:/ as the most confusable. Interestingly, asymmetrical confusions are observed with very few short vowels being misperceived as long vowels, but a larger number of long vowels misperceived as short. Consistent with previous studies of perception of English vowels in white noise [e.g., Benki (2003)], the findings confir...

Acoustic and perceptual effects of vowel length on voice onset time in Thai stops

In Thai, vowel length (e.g., [pan]‐[paan]) and unaspirated–aspirated stop (e.g., [pan]‐[phan]) distinctions are contrastive; this study examines possible acoustic and perceptual interactions between these temporal properties. VOT and vowel duration were measured for five repetitions of 40 (near‐) minimal word pairs beginning with aspirated or unaspirated bilabial, alveolar, or velar stops followed by long or short vowels, produced by three native Thai speakers. Although VOT for velars was slightly longer (about 5 ms) before long than before short vowels, the overall acoustic picture across places of articulation showed no systematic effect of vowel length. To test for perceptual influences of contrastive vowel length on VOT, a 15‐step 0–90 ms VOT continuum was created by editing the aspiration portion of naturally produced [phaan]. The VOT continuum was spliced onto the vowel portion of four different natural tokens, [pan], [paan], [phan], and [phaan], creating four continua, two with long vowels and two ...

Development of a Thai phonetically balanced monosyllabic word recognition test: Derivation of phoneme distribution, word list construction, and response evaluations

Abstract This paper proposes a test tool for Thai word recognition, the Thammasat University Phonetically Balanced Word List 2014 (TU PB’14), standardized on several major criteria: phonemic balance, familiarity, reliability, list equivalency, and homogeneity. Phoneme distributions from the largest written Thai corpus (InterBEST) were obtained and used to construct five phonetically balanced word lists, each containing 25 frequently occurring monosyllabic words. Listeners’ percent correct discrimination scores from test and re-test sessions were not significantly different, confirming test reliability. Detailed analysis of listeners’ errors revealed that perceptual errors occurred predominantly for initial sound only, final only, and initial together with final. In terms of list equivalency and homogeneity, derived psychometric function slopes of TU PB’14 ranged from 0.0941 to 0.1155, while intensities required for 50% intelligibility ranged from 41.0279 to 41.3697. Two-way Chi-Square analysis performed on both parameters indicated that there was no significant difference among the word lists.

Role of carrier signal types in perception of Thai phonemes: Implications for cochlear implant recipients

It is well known that segmental (consonants and vowels) and suprasegmental (e.g., tone and stress) speech sounds provide intrinsically different kinds of perceptual information. These differences suggest that phoneme perception might be improved in cochlear implant strategy by employing a carrier signal that is most compatible to each type of speech sounds. However, all speech vocoder strategies for cochlear implant (CI) available in most of works use one carrier signal due to their simplicity to implement. In this paper, we investigated the role of five different carrier signals for Thai speech perception on different types of phonemes, i.e., sine-carrier with/without band-pass filter (TF/TNF), noise-carrier with/without band-pass filter (NF/NNF), and temporal fine structure carrier (TFS). Each type of carrier signal was used to synthesize speech stimuli using the continuous interleaved sampling (CIS) strategy as basic and commonly used for CI. Four different psychoacoustic tests for initials, finals, vo...

Durational and spectral differences in Thai diphthongs and final glides

Acoustic analysis was conducted to compare Thai monophthongs /i, ii, Ɯ, ƜƜ, u, uu, a, aa/, diphthongs /ia, Ɯa, ua/, and vowel-to-glides (vowel + /j/or /w/) in terms of duration, formant frequency, and spectral rate of change (TLroc). Preliminary results from multiple repetitions of 30 target monosyllabic words (from 3 males) show that durational values of long monophthongs are approximately two times as long as the short counterparts. The diphthong onsets (/i/, /Ɯ/, /u/) and offsets (/a/) as part of diphthongs appear to be more centralized than when they occur in monophthongs. Syllables with final glides appear to be longer than those with other final consonants, reducing length differences between syllables with short and long vowels. Interestingly, not only that Thai diphthongs and vowel-to-glides differ in their articulatory trajectories, but they appear to differ in terms of duration and TLroc values. Average duration of diphthongs is shorter than that of vowel-to-glides. TLroc of diphthongs is on ave...

Phonetically balanced and psychometrically equivalent monosyllabic word lists for word recognition testing in Thai

In speech audiometry, a common and crucial method to obtain a suprathreshold (dB) at which words are repeated with maximum accuracy is referred to as word/speech recognition testing. For Thai, Thammasat University and Ramathibodi Hospital Phonetically Balanced Word Lists 2015 (TU-RAMA PB’15) were created with five lists, each with 25 monosyllabic words. Besides its phoneme distributions being based on large-scale Thai spoken corpora [1], TU-RAMA PB’15 is in line with TU PB’14 [2], [3] with emphasis on phonetic balance, symmetrical phoneme occurrence, and word familiarity. To evaluate its homogeneity in terms of decibel intelligibility, the lists were recorded and presented to 10 normal hearing participants, ranging from 0 to 50 dB HL in 2 dB increments (ascending order) until they repeated correct verbal responses. Using logistic regression, regression slopes and intercepts were calculated to estimate percentage of correct performance at any given intensity and to construct psychometric functions for ever...

Thai phonetically balanced word recognition test: Reliability evaluations and bias and error analysis

Word recognition score (WRS) is one of the measuring techniques used in speech audiometry, a part of a routine audiological examination. The test’s accuracy is crucial and largely depends on the test materials. With emphasis on phonetic balance, test-retest reliability, inter-list equivalency, and symmetrical phoneme occurrence, Thammasat University Phonetically Balanced Word Lists 2014 (TU PB’14) were created with five different lists, each with 25 Thai monosyllabic words. TU PB’14 reflects Thai phoneme distribution based on large-scale written Thai corpora, InterBEST [1]. To evaluate its validity and test-retest reliability, the lists were given at five intensity levels (15–55 dB HL) in test and retest sessions to 30 normal-hearing subjects. The differences in performance between the two sessions are not significantly large and correlation coefficients at the linear regions are all positive. Analysis of listeners’ errors, including sequence recurrences, was carried out. Errors occurred predominantly in ...

Markov random field in speech enhancement: Application for tonal languages

This paper proposed speech enhancement algorithm based on Markov random field (MRF) model for Thai, a tonal language. Firstly, a noisy speech signal is transformed using the short time Fourier transform (STFT). In so doing, noise is removed and speech is preserved, especially harmonics information as f0 patterns are relevant perceptual cues for lexical tones. The voice activity detector is used to classify each STFT time frame into voiced and unvoiced. Harmonics information is retrieved from each voiced time frame, where four neighborhoods of the analyzed STFT coefficients include its adjacent time frames (left, right) and nearest harmonics (top, bottom). For the unvoiced, four adjacent coefficients (left, right, top, and bottom) are used. A two-state MRF model is used to classify STFT coefficients into speech and noise. Those with speech state are retained, while the rest is set to zero. The enhanced speech is estimated by the inverse STFT. Results from quality evaluation test on four sets of Thai rhymin...