Algoritmos Genéticos Aplicados para Otimização dos Parâmetros de um Reconhecedor Automático de Fala

State-of-the-art automatic speech recognition (ASR) systems are comprised of complex decoders that have a large number of parameters that allow tuning for performance and speed. These parameters are often optimized manually, based on past experience and specialist knowledge. Specically, to reach optimal performance, deep understanding of each decoder parameter is necessary, as well as an understanding about the tradeo between accuracy word rate (AWR) and real time factor (RTF). In order to nd the optimal decoding parameters without an manual tuning, this paper presents a strategy that automatically tunes the Julius decoder parameters, used as engine in an ASR system. Particularly, the proposed strategy uses genetic algorithms to reach an optimal tune by evaluation of a cost function involving AWR and RTF. The obtained results of AWR and RTF are presented conrming the eectiveness of the proposed strategy.