| Paper: | SP-P13.12 | ||
| Session: | General Topics in Robust Speech Recognition | ||
| Time: | Thursday, May 20, 13:00 - 15:00 | ||
| Presentation: | Poster | ||
| Topic: | Speech Processing: Robust Speech Recognition | ||
| Title: | EXTENDED CLUSTER INFORMATION VECTOR QUANTIZATION (ECI-VQ) FOR ROBUST CLASSIFICATION | ||
| Authors: | Jon Arrowood; Nexidia, Inc. | ||
| Mark Clements; Georgia Institute of Technology | |||
| Abstract: | This paper presents a novel extension to vector quantization referred to asit Extended Cluster Information (ECI). In this method the decoder retainsmore general statistics about the VQ clusters found during codebook trainingthan the single prototypical point of conventional VQ systems. Typicallythis information is unnecessary, however if the items being compressed arefeature space vectors used as input to a statistical pattern classificationsystem, the extra probabilistic information can be used during theclassification as in Bayes Predictive Classification (BPC) to improverecognition results. To demonstrate one potential use of ECI-VQ, the Aurora2distributed speech recognition front end is altered to provide moreaggressive Mel Frequency Cepstral Coefficient (MFCC) compression. As thebit-rate drops, the corresponding recognition performance suffers. It isthen shown that using ECI-VQ as the input to an Uncertain Observation (UO)speech recognizer, a significant number of errors due to compression can becorrected with no extra cost in bit-rate. | ||
| Back | |||
Home -||-
Organizing Committee -||-
Technical Committee -||-
Technical Program -||-
Plenaries
Paper Submission -||-
Special Sessions -||-
ITT -||-
Paper Review -||-
Exhibits -||-
Tutorials
Information -||-
Registration -||-
Travel Insurance -||-
Housing -||-
Workshops