Implementing neural networks into modern technology
M. Johnson · 2003
The practical implementation of signal separation methods such as artificial neural networks and independent component analysis (ICA) bring several problems to light. These problems include time independence, stationarity, linearity, prior knowledge of sources, and hardware implementation. A direct emphasis is placed on implementing these networks on vocal signals, more specifically, voice-activated systems (VAS). What aspects and limitations of ICA and BSS (blind source separation) must be considered are given; the significance of preprocessing or prewhitening is considered; and learning algorithms are given. Also, a discussion is given on hardware enhancement, which may solve some of the problems related to ICA.