Method for detecting the best path through a stochastic network, in particular for speech or image recognition.

摘要

Method for determining the most likely path through a stochastic network such as a Markov model, with various subnetworks such as network branches. Instead of tracing the best path from the end state (backtracing), there is detected in all cases for at least one state, through which of those different subnetworks runs the most likely path ending in that state, and this information is propagated through the network. It is possible to assign to each subnetwork an identification code which is representative of that subnetwork and which is passed through the network to the end state. The method can be used in particular in the recognition of speech or image signals and the like.