VOICE RECOGNITION BY NEURAL-NETWORK UNDER TRACTOR NOISE

Experiments were conducted to investigate the possibility of using a neural network for operator's voice recognition under tractor noise. This concept will be used to control some of the tractor's functions by oral commands from the operator. Due to limited capacity of the network used, this investigation focused on the recognition of the vowels a, e, i, o, and u. The effect of variation in pitch of input voice and input units was investigated. The neural network was initially trained to recognize input voice and noise, and later during recognition, it compared the learned signals and test signals to distinguish between the tractor noise and operator's voice. All investigations were carried out at a fixed tractor noise level at 2500 rev/min of the engine. The effect of variation in sound pressure level was studied. It was observed that the neural network can be successfully used to recognize the operator's voice under the tractor noise. Input units of 256 or 128 were sufficient to recognize the steady state vowels used in this study. The sound pressure level affected the misrecognitions.