Optimization of Observation Membership Function By Particle Swarm Method for Enhancing Performances of Speaker Identification

The performance of speaker identification is severely degraded in noisy environments. Kim and et al suggested the concept of observation membership for enhancing performances of speaker identification with noisy speech [1]. The method is to weight observation probabilities with observation membership values decided by SNR. In the paper [1], the authors suggested heuristic parameter values for observation membership. In this paper we apply particle swarm optimization (PSO) for obtaining the optimal parameters. With the speaker identification experiments using the VidTimit database we verify that the optimization approach can achieve better performances than the original membership function. Key-Words: Speaker identification, observation membership, particle swarm optimization