The Role of Sound Change in the Speech Recognition System: A Phonetic Analysis of the Final Nasal Shift in Mandarin

The Role of Sound Change in the Speech Recognition System: A Phonetic Analysis of the Final Nasal Shift in Mandarin

過去十幾年來，電腦語言學家一直嘗試要設計一個能辨識標準語的語音系統，並能兼顧個別使用者的語音差異。目前，一些語音辨識系統已能針對使用者的年齡和性別，處理不同的音頻問題。然而，語音變化的問題還沒納入考慮。因此本文提議將語音變化的現象，整合到語音辨識系統的設計。本文以中文的字尾鼻音變化為例，分析它的語音特質，並討論它對語音辨識系統的影響，最後提出一個能提升辨識字尾鼻音的解決方案。

Over the past decade, computational linguists have been striving to design a speech recognition system that is able to identify standard speeches and to accommodate sound variables caused by different individual accents. Furthermore, some speech recognition programs have been able to learn and identify distinctive sound frequencies due to the user's age and gender. Nevertheless, regular sound alterations that occur in different varieties of a language have never been seriously considered in the design of the speech recognition system. Accordingly, this study proposes to incorporate the socio-phonological information about regular sound modifications to enhance the performance of Automatic Speech Recognition. To illustrate this point, this study investigates and analyzes the acoustic variation of the syllable-final nasal shift from the velar to the dental, which has been discovered to be one of the distinctive sound features that makes the variety of Mandarin spoken in Taiwan differ from that spoken in China. Following the phonetic analysis, this study discusses the effect of the nasal merger on the development of phonology-dependent speech technologies. It concludes by proposing a preliminary resolution to the identification of syllable-final nasals for the design of Automatic Speech Recognition.