Speech processing with a Boltzmann machine

Recent work on the representation of 'knowledge' in massively parallel networks has resulted in the development of the Boltzmann Machine. In such networks, 'knowledge' is represented by the pattern and strengths of interconnections between simple processing elements. In the Boltzmann Machine noise is used to aid the search procedure in the network. Developed primarily for static image processing, Boltzmann Machines are well suited to problems that can be formulated as constraint satisfaction searches. This paper describes some simple experiments using a Boltzmann Machine to process and classify static speech patterns.