Knowledge Resources in Automatic Speech Recognition and Understanding for Romanian Language
Inge Gavăt, Diana Mihaela, Corneliu Octavian · InTech eBooks · 2008
Chair of Applied Electronics and Information Engineering, Faculty of Electronics, Telecommunications and Information Technology.Our research interests concern mainly ASRU, but also other topics like speech recognition based on neural networks (Valsan et al., 2002), (Gavat & Dumitru, 2002-1), (Gavat et al., 2002-2), (Dumitru & Gavat, 2008) on fuzzy technics (Gavat & Zirra, 1996-2), (Gavat et al., 1997), (Gavat et al., 2001-1), (Gavat et al., 2001-2), or on Support Vector Machines (SVM) (Gavat et al., 2005-2), (Gavat & Dumitru, 2008), speech synthesis, TTS systems, speech and music retrieval (Gavat et al., 2005-1), speech prosody, multimodal systems, could be mentioned.Our main realization is the Automatic Speech Recognition System for Romanian Language, ASRS_RL (Dumitru, 2006), a research platform in order to implement and enhance different methods for ASRU in Romanian language (Gavat and Dumitru, 2008).Recognizers for phonemes, digits and continuous speech acting under the statistic paradigm based on hidden Markov models, under the connectionist paradigm of artificial neural networks or under fuzzy principles, but also under combined principles were experimented.The system presented in this chapter has as main objective refinement of the statistic paradigm for ASRU in Romanian language, by enhancement of two important aspects, acoustical modeling and language modeling.