Speech recognition by neural network adapted to reference pattern learning

Assignee: NEC Corp

•

Filing: 1992-04-23

•

Granted: —

•

Expires: 2012-04-23

📄 Original PDF 🌐 Google Patents 🔄 Sync Now

Abstract

Abstract A speech recognition method according to the present invention uses distances calculated through a variance weighting process using covariance matrixes as the local distances (prediction residuals) between the feature vectors of input syllables/sound elements and predicted vectors formed by different statuses of reference neural prediction models (NPM's) using finite status transition networks. The category to minimize the accumulated value of these local distances along the status transitions of all the prediction models is figured out by dynamic programming, and used as the recognition output. Learning of the reference prediction models used in this recognition method is accomplished by repeating said distance calculating process and the process to correct the parameters of the different statuses and the covariance matrixes of said prediction models in the direction of reducing the distance between the learning patterns whose category is known and the prediction models of the same category as this known category, and what have satisfied prescribed conditions of convergence through these calculating and correcting processes are determined as reference pattern models.

Inventors

👤 Ken-Ichi C/O Nec Corporation Iso

Bibliographic Details

Application Number —

Jurisdiction EP

Data Source google_patents

Last Synchronized 2026-11-01 22:34

CPC Classifications

G; G06; G06N; G06N3/00; G06N3/02; G06N3/04; G06N3/049; G06F; G06F18/00; G06F18/20; G06F18/29; G06F18/295; G10; G10L; G10L15/00; G10L15/08; G10L15/16

Claims Analysis

Total Claims: 4

Claim 1

Independent Claim 165 words

A pattern recognition method for recognizing syllables and sound elements on the basis of comparison of input time series patterns expressed as the feature vectors of the syllables and sound elements with reference pattern models using a finite status transition network, wherein each status of said finite status transition network has a predictor for calculating a predicted vector at time t from a plurality of feature vectors of said input time series patterns at time t-1 and before and a plurality of feature vectors at time t+1 and after; and a prediction residual determined by said input feature vector, said predicted feature vector by the predictor of said n -th status at time t , corresponding to said feature vector, and a covariance matrix accompanying said n th status is used as the local distance between the feature vector of said input time series patterns at time t , i.e. the input feature vector, and the n -th status of said finite status transition network.

Claim 2

Dependent Claim 61 words

A speech recognition method, as claimed in Claim 1, wherein correspondences between said input time series patterns and said predictors are so figured out by dynamic programming as to minimize the accumulated value of said local distances along said status transitions, and said accumulated value is used as the distance between said input time series patterns and said reference pattern models.

Claim 3

Dependent Claim 40 words

A speech recognition method, as claimed in Claim 2, wherein said accumulated value is calculated for every category of words, and the category having the smallest of said accumulated values which have been calculated is used as the recognition output.

Claim 4

Dependent Claim 108 words

A speech recognition method, as claimed in Claim 2 or 3, wherein initial values are set for the parameters of said predictor and said covariance matrix accompanying each status of said finite status transition network, said distance between said input time series pattern for the learning purpose, said category of which is known, and said reference pattern model corresponding to the same category as said known category is calculated; the parameters of said predictor and said covariance matrix of each state are iteratively corrected in the direction of reducing said distance; and said reference pattern model said distance for which satisfies predetermined conditions of convergence is thereby obtained.

References Cited (Backward Citations)

No citation records indexed.

USPTO Annuity & Maintenance Windows

Maintenance schedule is calculated for granted US utility patents.

Assigned Categories & Tags

No categories assigned to this patent yet.

Assign Category Tag

Select Category

Notes (Optional)