High-performance low-complexity wordspotting using neural networks

Author

Chang, Eric I. ; Lippmann, Richard P.

Author_Institution

Nuance Commun., Menlo Park, CA, USA

Volume

45

Issue

11

fYear

1997

fDate

11/1/1997 12:00:00 AM

Firstpage

2864

Lastpage

2870

Abstract

A high-performance low-complexity neural network wordspotter was developed using radial basis function (RBF) neutral networks in a hidden Markov model (HMM) framework. Two new complementary approaches substantially improve performance on the talker-independent Switchboard corpus. Figure of merit (FOM) training adapts wordspotter parameters to directly improve the FOM performance metric, and voice transformations generate additional training examples by warping the spectra of training data to mimic across-talker vocal tract length variability

Keywords

feedforward neural nets; hidden Markov models; learning (artificial intelligence); speech processing; speech recognition; FOM performance metric; HMM; across-talker vocal tract length variability; figure of merit training; hidden Markov model; high performance wordspotting; low complexity wordspotting; neural network wordspotter; radial basis function networks; spectra warping; talker-independent Switchboard corpus; training data; voice transformations; wordspotter parameters; Acoustic signal detection; Control systems; Hidden Markov models; Maximum likelihood detection; Measurement; NIST; Neural networks; Speech; Testing; Training data;

fLanguage

English

Journal_Title

Signal Processing, IEEE Transactions on

Publisher

ieee

ISSN

1053-587X

Type

jour

DOI

10.1109/78.650114

Filename

650114