Nuance Communications, Inc.

Kevin R Farrell

William F Ganong, Iii

David Alun James

Daniele Ernesto Colibro

Claudio Vair

Jerry K Carter

Almog Aley-Raz

Oshrit Yaron

Luca Scarpato

Pietro Laface

Sandro Cumani

Emanuele Dalmasso

Embodiments of the present invention perform speaker identification and verification by first prompting a user to speak a phrase that includes a common phrase component and a personal identifier. Then, the embodiments decompose the spoken phrase to locate the personal identifier. Finally, the embodiments identify and verify the user based on the results of the decomposing.

Method and apparatus for performing speaker recognition

Typical speaker verification systems usually employ speakers' audio data collected during an enrollment phase when users enroll with the system and provide respective voice samples. Due to technical, business, or other constraints, the enrollment data may not be large enough or rich enough to encompass different inter-speaker and intra-speaker variations. According to at least one embodiment, a method and apparatus employing classifier adaptation based on field data in a deployed voice-based interactive system comprise: collecting representations of voice characteristics, in association with corresponding speakers, the representations being generated by the deployed voice-based interactive system; updating parameters of the classifier, used in speaker recognition, based on the representations collected; and employing the classifier, with the corresponding parameters updated, in performing speaker recognition.

Method and apparatus for automated speaker parameters adaptation in a deployed speaker verification system

Techniques for automatically identifying a speaker in a conversation as a known person based on processing of audio of the speaker's voice to extract characteristics of that voice and on an automated comparison of those characteristics to known characteristics of the known person's voice. A speaker segmentation process may be performed on audio of the conversation to produce, for each speaker in the conversation, a segment that includes the audio of that speaker. Audio of each of the segments may then be processed to extract characteristics of that speaker's voice. The characteristics derived from each segment (and thus for multiple speakers) may then be compared to characteristics of the known person's voice to determine whether the speaker for that segment is the known person. For each segment, a degree of match between the voice characteristics of the speaker and the voice characteristics of the known person may be calculated.

Speaker verification methods and apparatus

A method for performing speaker recognition comprises: estimating respective uncertainties of acoustic coverage of respective speech utterance(s) by first and second speakers, the acoustic coverage representing respective sounds used by the speakers when speaking; representing the respective uncertainties of acoustic coverage in a manner that allows for efficient memory usage by discarding dependencies between uncertainties of different sounds for the speakers; representing the respective uncertainties of acoustic coverage in a manner that allows for efficient computation by representing an inverse of the respective uncertainties of acoustic coverage and then discarding the dependencies between the uncertainties of different sounds for the speakers; and computing a score between the speech utterance(s) by the speakers in a manner that leverages the respective uncertainties of the acoustic coverage during the comparison, the score being indicative of a likelihood that the speakers are the same speaker.

Fast speaker recognition scoring using I-vector posteriors and probabilistic linear discriminant analysis

Reliable speaker-based clustering of speech utterances allows improved speaker recognition and speaker-based speech segmentation. According to at least one example embodiment, an iterative bottom-up speaker-based clustering approach employs voiceprints of speech utterances, such as i-vectors. At each iteration, a clustering confidence score in terms of Silhouette Width Criterion (SWC) values is evaluated, and a pair of nearest clusters is merged into a single cluster. The pair of nearest clusters merged is determined based on a similarity score indicative of similarity between voiceprints associated with different clusters. A final clustering pattern is then determined as a set of clusters associated with an iteration corresponding to the highest clustering confidence score evaluated. The SWC used may further be a modified SWC enabling detection of an early stop of the iterative approach.

Method and apparatus for automatic speaker-based speech clustering

One aspect includes determining validity of an identity asserted by a speaker using a voice print associated with a user whose identity the speaker is asserting, the voice print obtained from characteristic features of at least one first voice signal obtained from the user uttering at least one enrollment utterance including at least one enrollment word by obtaining a second voice signal of the speaker uttering at least one challenge utterance that includes at least one word not in the at least one enrollment utterance, obtaining at least one characteristic feature from the second voice signal, comparing the at least one characteristic feature with at least a portion of the voice print to determine a similarity between the at least one characteristic feature and the at least a portion of the voice print, and determining whether the speaker is the user based, at least in part, on the similarity.

In one aspect, a method for determining a validity of an identity asserted by a speaker using a voice print is provided. The method comprises acts of performing a first verification stage comprising comparing a first voice signal from the speaker uttering at least one first challenge utterance with at least a portion of the voice print and performing a second verification stage if it is concluded in the first verification stage that the first voice signal was obtained from an utterance by the user. The second verification stage comprises adapting at least one parameter of the voice print based, at least in part, on the first voice signal to obtain an adapted voice print, and comparing a second voice signal from the speaker uttering at least one second challenge utterance with at least a portion of the adapted voice print.

In one aspect, a method for determining validity of an identity asserted by a speaker using a voice print associated with a user whose identity the speaker is asserting, the voice print obtained from characteristic features of at least one first voice signal obtained from the user uttering at least one enrollment utterance including at least one enrollment word is provided. The method comprises acts of obtaining a second voice signal of the speaker uttering at least one challenge utterance, wherein the at least one challenge utterance includes at least one word that was not in the at least one enrollment utterance, obtaining at least one characteristic feature from the second voice signal, comparing the at least one characteristic feature with at least a portion of the voice print to determine a similarity between the at least one characteristic feature and the at least a portion of the voice print, and determining whether the speaker is the user based, at least in part, on the similarity between the at least one characteristic feature and the at least a portion of the voice print.

Growing community of inventors

Medford, MA, United States of America

Kevin R Farrell