Back to results
US 12,592,238 B2Granted
Information processing method, information processing device, and non-transitory computer readable recording medium storing information processing program
AssigneePanasonic Intellectual Property Corporation of America
InventorKousuke Itakura
Life of this patent
Estimating the term…
Abstract
A speaker recognition device acquires a registered voice, converts the acquired registered voice to a plurality of property converted voices having respective acoustic properties different from each other, extracts a speaker feature indicative of a characteristic of a speaker from the registered voice, extracts a speaker feature from each of the property converted voices, compares all pairs of speaker features of a part or all of the speaker feature extracted from the registered voice and the speaker features extracted from the property converted voices, and calculates a threshold used for recognition of a speaker of an input voice on the basis of a result of the comparison.
Claims11
An information processing method, by a computer, comprising:
- acquiring a registered voice;
- converting the acquired registered voice to a plurality of property converted voices having respective acoustic properties different from each other;
- extracting a speaker feature indicative of a characteristic of a speaker from the registered voice;
- extracting a speaker feature from each of the property converted voices;
- comparing all pairs of speaker features of a part or all of the speaker feature extracted from the registered voice and the speaker features extracted from the property converted voices;
- and
- calculating a threshold used for recognition of a speaker of an input voice on the basis of a result of the comparison;
- and wherein, in the comparison of all the pairs of speaker features, a similarity between each of all the pairs of speaker features of the part or all of the speaker feature extracted from the registered voice and the speaker features extracted from the property converted voices is calculated, and
- in the calculation of the threshold, a cumulative distribution function F (x) which indicates a probability of a similarity having a value x or smaller in the calculated similarities is calculated;
- and a value of a similarity which corresponds to a predetermined probability of the calculated cumulative distribution function F (x) is calculated as the threshold on the basis of the calculated similarities.
7 more elements
An information processing device, comprising:
- an acquisition part that acquires a registered voice;
- a conversion part that converts the acquired registered voice to a plurality of property converted voices having respective acoustic properties different from each other;
- a first extraction part that extracts a speaker feature indicative of a characteristic of a speaker from the registered voice;
- a second extraction part that extracts a speaker feature from each of the property converted voices;
- a comparison part that compares all pairs of speaker features of a part or all of the speaker feature extracted from the registered voice and the speaker features extracted from the property converted voices;
- and
- a calculation part that calculates a threshold used for recognition of a speaker of an input voice on the basis of a result of the comparison and wherein, in the comparison of all the pairs of speaker features, a similarity between each of all the pairs of speaker features of the part or all of the speaker feature extracted from the registered voice and the speaker features extracted from the property converted voices is calculated, and
- in the calculation of the threshold, a cumulative distribution function F (x) which indicates a probability of a similarity having a value x or smaller in the calculated similarities is calculated;
- and
- a value of a similarity which corresponds to a predetermined probability of the calculated cumulative distribution function F (x) is calculated as the threshold on the basis of the calculated similarities.
7 more elements
A non-transitory computer readable recording medium storing an information processing program causing a computer to:
- acquire a registered voice;
- convert the acquired registered voice to a plurality of property converted voices having respective acoustic properties different from each other;
- extract a speaker feature indicative of a characteristic of a speaker from the registered voice;
- extract a speaker feature from each of the property converted voices;
- compare all pairs of speaker features of a part or all of the speaker feature extracted from the registered voice and the speaker features extracted from the property converted voices;
- and
- calculate a threshold used for recognition of a speaker of an input voice on the basis of a result of the comparison and wherein, in the comparison of all the pairs of speaker features, a similarity between each of all the pairs of speaker features of the part or all of the speaker feature extracted from the registered voice and the speaker features extracted from the property converted voices is calculated, and
- in the calculation of the threshold, the threshold is calculated a cumulative distribution function F (x) which indicates a probability of a similarity having a value x or smaller in the calculated similarities is calculated;
- and
- a value of a similarity which corresponds to a predetermined probability of the calculated cumulative distribution function F (x) is calculated as the threshold on the basis of the calculated similarities.
7 more elements
Dependencies are inferred from the claim text.