A Pruning Method for Reducing Calculation Costs of Speaker Identification System

概要

論文の詳細を見る
GMM (Gaussian Mixture Model) based speaker identification systems using ML (Maximum Likelihood) and WMR (Weighting Model Rank) demonstrate very high performances. However, such systems are not so effective in real use because of their high calculation costs. In this report, we propose a new pruning algorithm for decreasing calculation cost. In the algorithm, we select 30% of speaker models having higher likelihood with a part of input speech and then apply MWMR (Modified Weighted Model Rank) to those speaker models to find out identified speaker. To test the effectiveness of the proposed algorithm, we performed speaker identification experiments using several databases including TIMIT. The proposed method showed more than 30% of reduced processing time than conventional system while maintaining recognition accuracy.
社団法人電子情報通信学会の論文
2003-04-18