Lexical Tone Recognition in Mizo using Acoustic-Prosodic Features

14044

Authors

nan

Abstract

AbstractMizo is an under-studied Tibeto-Burman tonal language of the North-East India. Preliminary research findings have confirmed that four distinct tones of Mizo (High, Low, Rising and Falling) appear in the language. In this work, an attempt is made to automatically recognize four phonological tones in Mizo distinctively using acoustic-prosodic parameters as features. Six features computed from Fundamental Frequency (F0) contours are considered and two classifier models based on Support Vector Machine (SVM) & Deep Neural Network (DNN) are implemented for automatic tonerecognition task respectively. The Mizo database consists of 31950 iterations of the four Mizo tones, collected from 19 speakers using trisyllabic phrases. A four-way classification of tones is attempted with a balanced (equal number of iterations per tone category) dataset for each tone of Mizo. it is observed that the DNN based classifier shows comparable performance in correctly recognizing four phonological Mizo tones as of the SVM based classifier.

Keywords

nan

Full Content

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6458–6461 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6458 Lexical Tone Recognition in Mizo using Acoustic-Prosodic Features Parismita Gogoi1,3, Abhishek Dey4, Wendy Lalhminghlui1, Priyankoo Sarmah1, S. R. M. Prasanna1,2 1Indian Institute of Technology Guwahati, 2Indian Institute of Technology Dharwad, 3Dibrugarh University, 4Kaliber.ai Guwahati-781039, Dharwad-580011, Dibrugarh-786004, Guwahati-781039, India {parismitagogoi, wendy, priyankoo, prasanna}@iitg.ac.in, {abhishekdey.gu}@gmail.com Abstract Mizo is an under-studied Tibeto-Burman tonal language of North-East India. Preliminary research findings have confirmed that four distinct tones of Mizo (High, Low, Rising, and Falling) appear in the language. In this work, an attempt is made to automatically recognize four phonological tones in Mizo distinctively using acoustic-prosodic parameters as features. Six features computed from Fundamental Frequency (F0) contours are considered, and two classifier models based on Support Vector Machine (SVM) & Deep Neural Network (DNN) are implemented for automatic tone recognition task respectively. The Mizo database consists of 31,950 iterations of syllables covering the four tones in Mizo, collected from 19 speakers using trisyllabic phrases. A four-way classification of tones is attempted with a balanced (equal number of iterations per tone category) dataset for each tone of Mizo. It is observed that the DNN-based classifier shows a performance in recognizing the four phonological tones in Mizo that is comparable to the SVM-based classifier. Keywords: Tone recognition, Mizo, Tibeto-Burman, Pitch, Tone language 1. Introduction The Mizo language belongs to the Kuki-Chin subgroup of Tibeto-Burman language family, and is spoken by 830, 846 (Census of India, 2011) people predominantly in Mizoram province in the Northeast of India. Mizo has four distinct lexical tones namely High (H), Low (L), Rising (R) and Falling (F) tone (Fanai, 1992; Chhangte, 1993; Sarmah and Wiltshire, 2010b). Two sets of tonal minimal pairs demon- strating the four-way contrast of Mizo tones obtained from (Lalhminghlui and Sarmah, 2018) are presented in Table 1. Previous acoustic studies on Mizo tones have shown that the four Mizo tones are distinct from each other in terms of F0 slope, averaged F0 and duration (Sarmah and Wilt- shire, 2010b; Sarmah et al., 2015). In terms of duration, the canonical Mizo Falling tone is significantly shorter than the other three tones (Sarmah and Wiltshire, 2010b). In terms of average F0, the four tones in Mizo may not be distinct. However, in terms of F0 slope, all the four tones are significantly distinct from each other (Sarmah and Wilt- shire, 2010b). As seen in Figure 1, the High and the Low tone in Mizo are comparatively static while the other tones have dynamic pitch contours. The Falling tone begins from nearly the same point of initiation of High tone and the Low tone has a falling contour. The Rising tone has an initial downward dip and then rises up from about 32% of the to- tal duration. The presence of the effect of intrinsic F0 on the four tones in Mizo is also reported where high vow- els imposed higher F0 …

Content truncated to 500 words

Labeled Text

AbstractMizo is an under-studied Tibeto-Burman tonal language of the North-East India. Preliminary research findings have confirmed that four distinct tones of Mizo (High, Low, Rising and Falling) appear in the language. In this work, an attempt is made to automatically recognize four phonological tones in Mizo distinctively using acoustic-prosodic parameters as features. Six features computed from Fundamental Frequency (F0) contours are considered and two classifier models based on Support Vector Machine (SVM) & Deep Neural Network (DNN) are implemented for automatic tonerecognition task respectively. The Mizo database consists of 31950 iterations of the four Mizo tones, collected from 19 speakers using trisyllabic phrases. A four-way classification of tones is attempted with a balanced (equal number of iterations per tone category) dataset for each tone of Mizo. it is observed that the DNN based classifier shows comparable performance in correctly recognizing four phonological Mizo tones as of the SVM based classifier.

Publication Details

Paper ID:
14044
Year:
2020
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan