Method and apparatus for speaker-calibrated speaker detection

Patent metadata

Field	Value
Publication number	US-9564134-B2
Application number	US-201514868226-A
Country	US
Kind code	B2
Filing date	Sep 28, 2015
Priority date	Dec 21, 2011
Publication date	Feb 7, 2017
Grant date	Feb 7, 2017

Abstract

The present invention relates to a method and apparatus for speaker-calibrated speaker detection. One embodiment of a method for generating a speaker model for use in detecting a speaker of interest includes identifying one or more speech features that best distinguish the speaker of interest from a plurality of impostor speakers and then incorporating the speech features in the speaker model.

First claim

The invention claimed is:

1. A method for generating a speaker model for use in detecting a speaker of interest, the method implemented by instructions embodied in one or more non-transitory computer accessible storage media and executable by a processor, the method comprising:
identifying one or more speech features that best distinguish the speaker of interest from a plurality of impostor speakers by:
obtaining a plurality of speech samples, the plurality of speech samples comprising a set of speech samples from the speaker of interest and a plurality of additional speech samples from the plurality of impostor speakers;
extracting a plurality of speech features from the plurality of speech samples; and
ranking the plurality of speech features according to an ability to distinguish the speaker of interest from the plurality impostor speakers,

wherein the ranking comprises:
modeling one or more speech features within one or more regions; and
assigning a performance measure to the one or more speech features subsequent to the modeling, wherein the performance measure for an associated one of the one or more speech features represents a strength with which the associated one of the one or more speech features accurately distinguishes the speaker of interest from the plurality of impostor speakers;

wherein at least one signal indicative of the identified one or more speech features is provided to a software application to generate the speaker model; and
detecting, by the software application, the speaker of interest based on the generated speaker model.

2. The method of claim 1 , wherein at least one of the one or more speech features comprises a combination of two or more individual speech features.

3. The method of claim 1 , wherein the incorporating comprises: assigning a weight to each of the one or more speech features, based on the performance measure associated with the each of the one or more speech features.

4. The method of claim 3 , wherein a weight of zero excludes an associated one of the one or more speech features from the speaker model.

5. The method of claim 3 , wherein the assigning is performed by a classifier.

6. The method of claim 1 , wherein the one or more speech features includes at least one of: a cepstral feature, a prosodic feature, or a signal processing-based feature.

7. The method of claim 6 , wherein the cepstral feature is constrained by one of: a lexical feature, a phonetic feature, a state-level feature, a prosodic feature, a pause feature, a turn feature, or a speaking-rate feature.

8. The method of claim 1 , wherein the one or more speech features vary from one speaker of interest to another speaker of interest.

Assignees

• Stanford Res Inst Int

Inventors

• Shriberg Elizabeth
• Ferrer Luciana
• Stolcke Andreas
• Graciarena Martin
• Scheffer Nicolas

Classifications

• G10L17/26 Primary

  Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices · CPC title

• G10L17/02 Primary

  Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction · CPC title

• G10L17/04

  Training, enrolment or model building · CPC title

• G10L17/24

  the user being prompted to utter a password or a predefined phrase · CPC title

• G10L17/06

  Decision making techniques; Pattern matching strategies · CPC title