arXiv AI recent: Improving Code-Switching ASR with Code-Mixing Guided Synthetic Speech
Researchers proposed a code-mixing guided preference-learning framework to improve code-switching Automatic Speech Recognition (ASR) using synthetic speech generation. The framework uses...
The experiments were conducted on the SEAME Mandarin-English conversational corpus, and the results show a reduction in MER from 12.1%/17.8% to 8.9%/14.2% on the DevMAN and DevSGE sets, respectively.