The SRI NIST 2010 Speaker Recognition Evaluation System (PREPRINT)

The SRI speaker recognition system for the 2010 NIST speaker recognition evaluation (SRE) incorporates multiple subsystemswith a variety of features and modeling techniques. We describe our strategy for this year's evaluation, from the use of speech recognition and speech segmentation to the in...

Full description

Saved in:
Bibliographic Details
Main Authors: Scheffer, Nicolas, Ferrer, Luciana, Graciarena, Martin, Kajarekar, Sachin, Shriberg, Elizabeth, Stolcke, Andreas
Format: Report
Language:English
Subjects:
ASR(AUTOMATIC SPEAKER RECOGNITION)
DATA FUSION
DCF(DECISION COST FUNCTIONS)
HMM(HIDDEN MARKOV MODELS)
Human Factors Engineering & Man Machine System
Linguistics
MARKOV PROCESSES
MATHEMATICAL MODELS
MESSAGE PROCESSING
MLLR(MAXIMUM LIKELIHOOD LINEAR REGRESSION)
NATURAL LANGUAGE
NIST(NATIONAL INSTITUTE OF STANDARDS AND TECHNOLOGY)
Numerical Mathematics
PROSODY
SENSOR FUSION
SPEAKER RECOGNITION
SPEECH RECOGNITION
SRE(SPEAKER RECOGNITION EVALUATION)
SYMPOSIA
SYSTEM FUSION
Voice Communications
VOICE RECOGNITION
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The SRI speaker recognition system for the 2010 NIST speaker recognition evaluation (SRE) incorporates multiple subsystemswith a variety of features and modeling techniques. We describe our strategy for this year's evaluation, from the use of speech recognition and speech segmentation to the individual system descriptions as well as the final combination. Our results show that under most conditions, the cepstral systems tend to perform the best, but that other, non-cepstral systems have the most complementarity. The combination of several subsystems with the use of adequate side information gives a 35% improvement on the standard telephone condition. We also show that a constrained cepstral systembased on nasal syllables tends to be more robust to vocal effort variabilities. To be presented at the International Conference on Acoustics, Speech, and Signal Processing (ICASSP) to be held in Prague, Czech Republic 22-27 May 2011. The original document contains color images.