隨著台灣逐漸步入高齡化社會，老年人口佔比也一年比一年還多，台語的地位也逐漸受到重視。為了與長輩溝通，有時會為了更正確解釋名詞或表達更生動而在句華語句子中摻入台語，或是講台語中因為想不起某個詞的台語而在台語句子中摻入華語。時常也有古早詞彙只有台語沒有華語，或是新穎詞彙只有華語沒有台語的狀況，對話中為了表達該物品或行為有時非得切換語言，形成了經常會出現國台夾雜對話的局面。
現今的主流辨識引擎(如 Google、Apple 的 Siri)無法辨識台語，更不用說是國台夾雜的句子。由於這些引擎沒有辦法達到實際應用的需求，因此我們決定自行開發國台夾雜的辨識引擎，並根據國台語音素上的異同作分析及記錄。有了這個國台辨識系統開發的經驗，未來也能將這相同的概念移轉到其他多語語音辨識系統上做使用。

As Taiwan has started to step into an aging society, the existing percentage of elders has been increasing year by year, and this leads to Taiwanese being more and more important. While interacting with elders in Taiwan, in order to express ourselves more clearly and correctly, sometimes we have to mix Taiwanese into our Mandarin sentences, or mix Mandarin into our Taiwanese sentences when we forget how a phrase in Taiwanese is pronounced. There's also situations when there are older phrases that don't have their Mandarin pronunciation or newer phrases that did not exist in the past, which means they don't have Taiwanese pronunciation to be spoken, therefore it's very common that there's Taiwanese-Chinese mixed sentences used everywhere.
The current ASR (automatic speech recognition) mainstreams are Google and Apple's Siri, but these two do not support Taiwanese ASR, let alone Taiwanese-Chinese mixed ASR. Due to the fact that these two ASRs cannot match the usage of ours, we decided to develop our own ASR system, and record differences in Taiwanese and Chinese language features. With this development experience, it can be implemented into our languages in the future as well.

[1] 內政部, “人口統計資料 - 中華民國內政部戶政司全球資訊網”，2023
[2] 教育部, “臺灣閩南語常用詞辭典”，2011
[3] Daniel Povey, Arnab Ghoshal, Gilles Boulianne, L. Burget, O. Glembek, N. Goel, M. Hannemann, P. Motlícek, Y. Qian, Petr Schwarz, J. Silovský, G. Stemmer, Karel Veselý, “The Kaldi Speech Recognition Toolkit”, 2011
[4] Taiwan AI Labs, “雅婷逐字稿”. 2018
[5] 教育部，”台灣閩南語羅馬字拼音方案使用手冊”，2007
[6] 教育部，”國語注音符號手冊”，2000
[7] Jyun-Ying Chen, “Effective ASR Error Detection & Correction System for Domain-Specific Task based on Context, syntactic & Semantic Information”, 2022
[8] Shuaibo Wang, Yufeng Chen, Songming Zhang, Deyi Xiong, Jinan Xu,“Adversarially Improving NMT Robustness to ASR Errors with Confusion Sets”,AACL, 2022
[9] Ahmed Ali, Steve Renals, “Word error rate estimation for speech recognition: e-WER,” ACL, 2018
[10] I. Vasilescu, D. Yahia, Natalie D. Snoeren, M. Adda-Decker, L. Lamel, “Cross-Lingual Study of ASR Errors: On the Role of the Context in Human Perception of Near-Homophones”, Interspeech, 2011
[11] Yi-Chang Chen, Chun-Yen Cheng, Chien-An Chen, Ming-Chieh Sung, YiRen Yeh,“Integrated Semantic and Phonetic Post-correction for Chinese Speech Recognition”, Taiwan Conference on Computational Linguistics and Speech Processing, 2021
[12] Satoshi Tsujioka, S. Sakti, Koichiro Yoshino, Graham Neubig, Satoshi Nakamura, “Unsupervised Joint Estimation of Grapheme-to-Phoneme Conversion Systems and Acoustic Model Adaptation for Non-Native Speech Recognition”, Interspeech, 2016
[13] Shao-Chuan Shen, “Taiwanese-Chinese Mixed Language Speech Recognition based on Multi-model Approach”, 2020