Noise Robust Automatic Speech Recognition

The recent advancements in automatic speech§recognition (ASR) §systems have expedited their commercial deployment.§Still, a major §challenge faced by the current ASR systems is their poor §performance for any mismatch between training and§testing §conditions. This book presents an in-depth analysis§of multi-§stream combination approach to improve the robustness§of ASR §systems for mismatch caused by additive noise. The§book also §discusses that multi-stream approach is complementary§to other §approaches for noise robustness, and how they can be§used §together. The selection of feature streams and§allocation of weights §to each feature stream are two important issues in§multi-stream. §Both the issues are analyzed, and a step-wise§approach is §presented to realize a significant improvement in§performance by a §multi-stream system compared to that of a§state-of-the-art §system. The analysis proposed in this §work can also be applied to other pattern recognition§tasks such as §speaker recogntion and text-processing, rendering§this book useful §for a wide audience ranging from beginner to advance§stage users §working in ASR as well as other pattern recognition§domains.