A Fixed-Point Neural Network Architecture for Speech Applications on Resource Constrained Hardware

摘要

Speech recognition and keyword detection are becoming increasingly popular applications for mobile systems. These applications have large memory and compute resource requirements, making their implementation on a mobile device quite challenging. In this paper, we design low cost neural network architectures for keyword detection and speech recognition. Wepresent techniques to reduce memory requirement by scaling down the precision of weight and biases without compromising on the detection/recognition performance. Experiments conducted on the Resource Management (RM) database show that for the keyword detection neural network, representing the weights by 5 bits results in a 6 fold reduction in memory compared to a floating point implementation with very little loss in performance. Similarly, for the speech recognition neural network, representing the weights by 6 bits results in a 5 fold reduction in memory while maintaining an error rate similar to a floating point implementation. Preliminary results in 40nm TSMC technology show that the networks have fairly small power consumption: 11.12mW for the keyword detection network and 51.96mW for the speech recognition network, making these designs suitable for mobile devices.