Bilingual Speech Recognition for Taiwanese Hokkien and Mandarin: Computational Intelligence in Complex Acoustic Environments
Jeng‐Shin Sheu, Aftab Ahmad, Jian-Min Li · 2024
This paper explores Taiwanese Hokkien, the second most spoken language in Taiwan, often intertwined with Mandarin. Existing Mandarin speech recognition struggles with the complexities of these languages' interaction, compounded by Taiwanese Hokkien's lack of a standardized writing system. Real-world challenges like ambient speech and indoor reverberation further hinder accuracy. To address these, we investigate methods for robust bilingual speech recognition, emphasizing tailored models. Additionally, we tackle consolidating textual representations, crucial given Taiwanese Hokkien's reliance on characters from other languages. This research aims to advance speech recognition for underrepresented languages like Taiwanese Hokkien, paving the way for more inclusive communication systems in diverse regions.