Recent improvements of the SpeeD Romanian LVCSR system

Author

Cucu, H. ; Buzo, Andi ; Petrica, Lucian ; Burileanu, Dragos ; Burileanu, C.

Author_Institution

Speech & Dialogue Res. Lab., Univ. Politeh. of Bucharest, Bucharest, Romania

fYear

2014

fDate

29-31 May 2014

Firstpage

1

Lastpage

4

Abstract

This paper presents the main improvements brought recently to the SpeeD automatic speech recognition system. Several aspects, such as speech and text resources acquisition, noise-robust speech features and feature transforms are discussed. All the updates in our ASR system are accompanied by experimental results illustrating significant improvements: between 30% and 35% relative WER reductions for various case studies (read/spontaneous speech, noisy/clean speech). In the last part of the paper, our ASR system is also compared with Google´s ASR system and a brief analysis of the results is presented.

Keywords

natural language processing; speech recognition; SpeeD Romanian LVCSR system; SpeeD automatic speech recognition system; feature transforms; noise robust speech features; speech resource acquisition; text resource acquisition; Google; Hidden Markov models; Noise measurement; Noise robustness; Speech; Speech recognition; ASR; LDA; LVCSR; PNCC; noise-robust speech recognition;

fLanguage

English

Publisher

ieee

Conference_Titel

Communications (COMM), 2014 10th International Conference on

Conference_Location

Bucharest

Type

conf

DOI

10.1109/ICComm.2014.6866659

Filename

6866659