Visible view data to reduce the multilingual gap on confidence models

The guiding reading of SSL (SSL) has made great progress in the learning of the talk. Models such as WAV2VECEC 2.0 and Hubert receive Kingdom results such as the recognition of speech recognition, especially in monolatial arrangements. However, many SSL languages ​​are often used to reduce their language in each language, especially in many languages ​​such as two languages ​​such as two languages ​​such as two languages. In this project, we investigate the novel system of reducing this gask to inform the visible site visible in the form of two languages ​​talk models. Our results indicate that the visual benefits of two languages ​​and languages, by reducing the multilingual gap in zero disposal from 31.5% in 8.04% models.

• 40 NAMPEETER University
• ** Work done while in Apple