LVCSR log-likelihood ratio scoring for keyword spotting

Author

Weintraub, Mitchel

Author_Institution

Speech Technol. & Res. Program, SRI Int., Menlo Park, CA, USA

Volume

1

fYear

1995

fDate

9-12 May 1995

Firstpage

297

Abstract

A new scoring algorithm has been developed for generating wordspotting hypotheses and their associated scores. This technique uses a large-vocabulary continuous speech recognition (LVCSR) system to generate the N-best answers along with their Viterbi alignments. The score for a putative hit is computed by summing the likelihoods for all hypotheses that contain the keyword normalized by dividing by the sum of all hypothesis likelihoods in the N-best list. Using a test set of conversational speech from Switchboard Credit Card conversations, we achieved an 81% figure of merit (FOM). Our word recognition error rate on this same test set is 54.7%

Keywords

maximum likelihood estimation; speech recognition; vocabulary; LVCSR system; N-best answers; N-best list; Switchboard Credit Card conversations; Viterbi alignments; conversational speech; figure of merit; hypothesis likelihoods; keyword spotting; large-vocabulary continuous speech recognition; log-likelihood ratio scoring; putative hit; scoring algorithm; test set; word recognition error rate; wordspotting hypotheses; Credit cards; Degradation; Distributed computing; Error analysis; Frequency; Hidden Markov models; High performance computing; Information systems; Management information systems; Resource management; Speech recognition; Testing; Viterbi algorithm;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1995. ICASSP-95., 1995 International Conference on

Conference_Location

Detroit, MI

ISSN

1520-6149

Print_ISBN

0-7803-2431-5

Type

conf

DOI

10.1109/ICASSP.1995.479532

Filename

479532