A hybrid NN-HMM system for connected digit recognition over telephone in Romanian language

Author

Gavat, Inge ; Zirra, Matei ; Enescu, Valentin

Author_Institution

Dept. of Electron. & Commun., Politehnica Univ. of Bucharest, Romania

fYear

1996

fDate

30 Sep-1 Oct 1996

Firstpage

37

Lastpage

40

Abstract

The paper presents a system for connected speech recognition with a limited number of words over the telephone line. The system, based on a hybrid structure (a hidden Markov model and an artificial neural net) takes into account the speaker and the telephone equipment variability by incorporating a preprocessing stage that subtracts the personal differences. Experiments performed on 100 speakers, with different system configurations: HMM, HMM+ANN, or with the entire system configuration, revealed an improvement of the error rate from 2% to 0.6% for the isolated digit recognition task and from 12% to 4% for connected strings

Keywords

errors; feedforward neural nets; hidden Markov models; multilayer perceptrons; speech recognition; telephony; Romanian language; artificial neural net; connected digit recognition; connected speech recognition; connected strings; error rate; hidden Markov model; multilayer perceptrons; personal differences; system configurations; telephone; telephone equipment; Artificial neural networks; Automatic speech recognition; Hidden Markov models; Humans; Natural languages; Paper technology; Speech enhancement; Speech processing; Speech recognition; Telephony;

fLanguage

English

Publisher

ieee

Conference_Titel

Interactive Voice Technology for Telecommunications Applications, 1996. Proceedings., Third IEEE Workshop on

Conference_Location

Basking Ridge, NJ

Print_ISBN

0-7803-3238-5

Type

conf

DOI

10.1109/IVTTA.1996.552717

Filename

552717