Segment-based speaker verification using dynamically generated phrases

Abstract

Methods, systems, and apparatus for a computer program for verifying the identity of a user, including a computer program encoded on a computer storage medium. The methods, systems, and apparatus include an act of receiving a request for a verification phrase for verifying a user's identity. Additional actions include: in response to receiving the request for the verification phrase used to verify the identity of the user, identifying subwords to be included in the verification phrase; The subwords included in the verification phrase, a candidate phrase including at least some of the identified subwords is obtained as the verification phrase. Additional actions include providing the verification phrase as a response to the request for the verification phrase, the verification phrase used to verify the identity of the user.

Description

Technical field

[0001] The present disclosure generally relates to speaker verification. Background technique

[0002] The computer can perform speaker verification to verify the identity of the speaker. For example, the computer may verify the identity of the speaker as the specific user based on verifying that the acoustic data representing the speaker's voice matches the acoustic data representing the voice of the specific user. Summary of the invention

[0003] In general, aspects of the subject matter described in this specification may involve a process for verifying the identity of a speaker. Speaker verification occurs by matching acoustic data representing utterances from the speaker with acoustic data representing utterances from a specific user.

[0004] The system can perform speaker verification by always asking the speaker to say the same phrase such as "FIXED VERIFICATION PHRASE." This method can be very accurate but can be easily deceived. For example, the record...

Examples

Embodiment Construction

[0024] figure 1 Is a flowchart of an example process 100 for verifying the identity of a speaker. In general, the process 100 may include a voice verification registration phase (110). For example, the system may prompt a specific user to speak a registered phrase and store training acoustic data indicating that the specific user uttered the registered phrase. For example, the acoustic data of each subword in the plurality of subwords may be the energy of the MFCC coefficient or the filter bank representing the specific user who spoke each subword in the plurality of subwords. The subword can be a phoneme or a sequence of two or more phonemes, such as a triphone. in figure 2 The registration stage of voice verification is illustrated in.

[0025] The process 100 may include a dynamic generation phase of verification phrases (120). For example, in response to a request for a verification phrase, the system can dynamically generate a verification phrase for verifying the identi...

Technical field

[0001] The present disclosure generally relates to speaker verification. Background technique

[0002] The computer can perform speaker verification to verify the identity of the speaker. For example, the computer may verify the identity of the speaker as the specific user based on verifying that the acoustic data representing the speaker's voice matches the acoustic data representing the voice of the specific user. Summary of the invention

[0003] In general, aspects of the subject matter described in this specification may involve a process for verifying the identity of a speaker. Speaker verification occurs by matching acoustic data representing utterances from the speaker with acoustic data representing utterances from a specific user.

[0004] The system can perform speaker verification by always asking the speaker to say the same phrase such as "FIXED VERIFICATION PHRASE." This method can be very accurate but can be easily deceived. For example, the record...

Examples

Embodiment Construction

[0024] figure 1 Is a flowchart of an example process 100 for verifying the identity of a speaker. In general, the process 100 may include a voice verification registration phase (110). For example, the system may prompt a specific user to speak a registered phrase and store training acoustic data indicating that the specific user uttered the registered phrase. For example, the acoustic data of each subword in the plurality of subwords may be the energy of the MFCC coefficient or the filter bank representing the specific user who spoke each subword in the plurality of subwords. The subword can be a phoneme or a sequence of two or more phonemes, such as a triphone. in figure 2 The registration stage of voice verification is illustrated in.

[0025] The process 100 may include a dynamic generation phase of verification phrases (120). For example, in response to a request for a verification phrase, the system can dynamically generate a verification phrase for verifying the identi...

Images