Tibetan Speech Recognition Based on the Am-do Dialect

What is it about?

In China, Tibetan is mainly divided into three dialects: Amdo, Khams, and Lhasa. The Amdo dialect evolved from ancient Tibetan, and while these dialects have developed differently, they all belong to the Tibetan language family. This study improved the pronunciation dictionary for the Amdo dialect and expanded speech data by data augmentation technology combining noise and reverberation to help computers recognize Amdo speech more accurately. Additionally, the language model was optimized based on Tibetan-specific grammar rules. Tests showed that the new method reduced error rates by about 3% and improved recognition accuracy by up to 19% compared to traditional approaches, delivering better results.

Featured Image