A reusable HMM soft-core for isolated word recognition

In this paper we introduce a scalable and reusable library component implementation of a HMM that may be used in order to be embedded in the full chain of an isolated word recognition system. Theoretically, the core allows for an unlimited /spl trade/scalability in the number of the vocabulary words (L) and in the number of symbols in the observation sequence of the word to be recognized (O). The number of hidden states (N) is also scalable, but limited to 16 and organized as a Bakis' model. The hardware design has been modeled in VHDL portable code resulting in a technology independent soft-core.