The primary advantages of the cerebellar model arithmetic computer (CMAC) are its ability to learn very fast and it can approximate a wide variety of non-linear functions. A comprehensive and efficient technique for speech enhancement based on an extension of the spectral subtraction method and integrating it with the higher order CMAC is developed. In addition, the paper also presents an unsupervised learning of the higher order CMAC as applied to speech enhancement. Simulation results using speech corrupted with very low signal to noise ratio (from -5dB to -20dB) in a vehicular environment using microphone placed on a dashboard in front of the speaker, shows great potential on the application of the HCMAC-ASS for practical application in signal enhancement.
展开▼