Embedded natural language processing for in-car speech commands


At a glance

Natural language processing (NLP) has recently become a core capability for many consumer devices. Combined with advances in speech recognition, NLP is critical for creating an intelligent, seamless experience for any product. This is especially true in safety-critical applications, such as in cars, when a lapse in attention due to a complex user interface can potentially result in a collision or even death. Moreover, a cloud based inference system may not be responsive for all driving situations. Recently, Deep Neural Networks (DNNs) have become widely used in NLP, having achieved state-of-the-art accuracy in essentially every domain, including user intent classification, question answering, and sentiment classification. The goal of this project is to bring the latest DNN-powered embedded NLP technologies to cars in a resource-efficient manner, enabling seamless in-car speech commands from passengers.

principal investigatorsresearchersthemes
Kurt KeutzerAmir Gholami

NLP, embedded deep neural networks, efficient inference