Noise modeling method to enhance speech recognition quality

Abstract

This invention proposes a noise modeling method to enhance speech recognition quality, enabling the recognition system to perform better in real-world environments. This method not only adds noise to the training signal to simulate different environments, but also incorporates the names of the noise types into the speech transcript. This allows the recognition model to better model different noise types and increase its accuracy.