ProtoLOGOS, system for Romanian language automatic speech recognition and understanding (ASRU)

Abstract

The ProtoLOGOS Romanian language automatic speech recognition and understanding system is based at the level of acoustical modeling on the statistical framework of hidden Markov models and at the linguistic level on a finite state grammar or a bigram language model. The system enhanced the speech signal modeling by using for the first time the PLP (Perceptual Linear Prediction) features in a Romanian language recognition system. The main performance indicator for our system is the phrase recognition rate. This paper presents the main results obtained using the two language models already mentioned and the METEO database, constituted by broadcasted forecast news records.