Modelos manejados por Sautrela

Sautrela defines two major interfaces to unify many different stochastic model techniques related to speech recognition: the deterministic and the non-deterministic Weighted Finite-state Automata (WFSA). Regardless of its internal structure, any model that implements one of them can be managed by the system: An integrated module takes advantage of the said interfaces and turns out to be a simple and general Decoder that implements well known search algorithms such as frame-synchronous Viterbi with configurable pruning.

However, a speech recognition system is typically composed of different models standing for the corresponding knowledge sources (language, lexical and acoustic knowledge). Hence, a single integrated stochastic formalism is needed, and Layered Markov Models (LMM) arise in response to this problem.

This topic: Sautrela > WebHome > ModelosManejadosPorSautrela
History: r4 - 08 Mar 2007 - 00:46:02 - MikelPenagarikano
Copyright Mikel PenagarikanoValid XHTML 1.0 Transitional