Improving Vietnamese Accent Recognition Using ASR Transfer Learning

Accent Recognition (AR) is a critical task in voice-controlled systems. If accent information is known in advance, voice-controlled systems can switch to a suitable accent-specific mode to improve their performance and user experience. However, available accent datasets, especially Vietnamese, are relatively small, making AR very challenging. To deal with these drawbacks, this paper proposes a transfer learning method using pretrained ASR models for Vietnamese accent recognition. This helps the system utilize available speech recognition systems while capturing implicit linguistic and phonetic information learned in ASR to improve its performance. Several experiments were conducted on a Vietnamese 8kHz telephone call dataset, which showed a significant improvement of the proposed system over existing Vietnamese AR models.