Automatic speech recognition (ASR) is becoming increasingly more integral in our daily lives. While recent breakthroughs have tremendously improved ASR performance, these models still suffer considerable degradation from ambient noise. Therefore, ASR robustness under adverse conditions becomes more important than ever. According to the processing stages of an ASR system, approaches for increasing ASR robustness can be classified into three groups: back-end, front-end, and joint training techniques. This thesis follows the aforementioned three axes of research.
«
Automatic speech recognition (ASR) is becoming increasingly more integral in our daily lives. While recent breakthroughs have tremendously improved ASR performance, these models still suffer considerable degradation from ambient noise. Therefore, ASR robustness under adverse conditions becomes more important than ever. According to the processing stages of an ASR system, approaches for increasing ASR robustness can be classified into three groups: back-end, front-end, and joint training techniqu...
»