Development of Deep Learning-based Speech Recognition Algorithm for FPGA

Automatic speech recognition (ASR)-based services have become an indispensable part of our daily life, thanks to products such as Apple Siri, Google Home, and Amazon Alexa. More and more smart home appliances are also controlled by voice. However, most of these services rely on speech recognition and natural language understanding services running in the cloud, but this introduces delay that sometimes is not acceptable. There are also some applications where there is no WiFi or cellular network to connect to the cloud. Therefore there is a need to develop a local speech recognition system with high performance. In this project, we will develop a deep learning neural network-based speech recognition algorithm using the latest technologies that are suitable for embedded systems, such as the Transformer. We will then work with the partner company to port the algorithm into a FPGA- based System on Chip (SoC) product.

Faculty Supervisor:

Jie Liang

Student:

Binglin Li

Partner:

Fraser Innovation Inc.

Discipline:

Engineering

Sector:

Information and cultural industries

University:

Simon Fraser University