A fast search method of speaker identification for large population using pre-selection and hierarchical matching

Zhibin Pan, Koji Kotani, Tadahiro Ohmi

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Performance of search during matching phase in a speaker identification system realized through vector quantization (VQ) is investigated in this paper. Voice of each person is recorded in a office room with personal computers. LPC-cepstrum is selected as feature vector. In order to gain higher success rate of identification, it is necessary to use larger size codebook for each person. Consequently, it is extremely time-consuming to do full matching directly with all of registered larger size codebooks for large population to determine who the current speaker is. To fulfill identification at a tolerable speed, trade-off between success rate and search time is unavoidable usually. In this paper, a fast search method is proposed that is based on pre-selection and hierarchical matching so as to eliminate a large amount of impossible candidates before doing final fine matching with larger size codebooks meanwhile keep the success rate not degraded. Pre-selection is achieved using divergences between unknown input speech and all registered codebooks. Hierarchical matching is implemented using smaller size codebook and larger size codebook corresponding to each person respectively. With this method, for a population of 290, time required for identification can be reduced to about 13% while memory occupation is increased by just 6% when baselines are taken as those using larger size codebooks in conventional way. Success rate is not degraded as 94% comparing to that acquired in conventional way as well.

Original languageEnglish
Title of host publication6th International Conference on Spoken Language Processing, ICSLP 2000
PublisherInternational Speech Communication Association
ISBN (Electronic)7801501144, 9787801501141
Publication statusPublished - 2000 Jan 1
Event6th International Conference on Spoken Language Processing, ICSLP 2000 - Beijing, China
Duration: 2000 Oct 162000 Oct 20

Publication series

Name6th International Conference on Spoken Language Processing, ICSLP 2000

Other

Other6th International Conference on Spoken Language Processing, ICSLP 2000
CountryChina
CityBeijing
Period00/10/1600/10/20

Keywords

  • Fast search
  • Hierarchical matching
  • Large population
  • Pre-selection

ASJC Scopus subject areas

  • Linguistics and Language
  • Language and Linguistics

Fingerprint Dive into the research topics of 'A fast search method of speaker identification for large population using pre-selection and hierarchical matching'. Together they form a unique fingerprint.

Cite this