ONLINE SPEAKER TESTING USING NEURAL-NETWORK-BASED SYSTEM

This paper presents a high-performance neural network-based system for the on-line testing of loud speakers. A multi-layer neural network sy stem with back-propagation learning algorithm is employed. It consists of 53 input nodes, one hidden layer with 10 nodes and 1 output node. The normalised Total Harmonics Distortion values of the speakers at di fferent frequencies are fed to the input of the system. The average tr aining time is 40 minutes (on a 486DX 50 MHz PC) for a training size o f 100 patterns. The neural network-based system is able to achieve a r emarkable accuracy of 95%.