DocumentCode :
1687792
Title :
Warped Minimum Variance Distortionless Response based bottle neck features for LVCSR
Author :
Kilgour, Kevin ; Tseyzer, Igor ; Quoc Bao Nguyen ; Waibel, Alex
Author_Institution :
Int. Center for Adv. Commun. Technol. - InterACT, Inst. for Anthropomatics, Karlsruhe, Germany
fYear :
2013
Firstpage :
6990
Lastpage :
6994
Abstract :
This paper presents the results of our experiments on bottleneck feature applied to a wMVDR (Warped Minimum Variance Distortionless Response) frontend. We examine how to best optimize wMVDR-BNF features and wMVDR combined with MFCC bottleneck features (wMVDR+MFCC-BNF). Our wMVDR+MFCC-BNF frontend improves a single pass system from 18.7% (20.7%) to 18.1% compared to a MFCC-BNF (MFCC) system tested on the Quaero 2010 German evaluation set. When used in a system combination our wMVDR-BNF and wMVDR+MFCC-BNF systems reduced the overall WER from 14.3% to 13.3% on the IWSLT 2010 test set while at the same time reducing the number of systems needed from 9 to 5. Our result of 11.9% on the 2012 IWSLT testset is better than the best result submitted during the evaluation campaign.
Keywords :
speech recognition; IWSLT 2010 test set; LVCSR; Quaero 2010 German evaluation set; WER; bottle neck features; bottleneck features; evaluation campaign; speech recognition; wMVDR+MFCC-BNF frontend; warped minimum variance distortionless response frontend; Adaptation models; Context; Mel frequency cepstral coefficient; Speech; Speech recognition; Training; ASR; BNF; MLP; Speech recognition; wMVDR;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on
Conference_Location :
Vancouver, BC
ISSN :
1520-6149
Type :
conf
DOI :
10.1109/ICASSP.2013.6639017
Filename :
6639017
Link To Document :
بازگشت