User verification method and apparatus using generalized user model

A speaker verification method comprising:

• generating a feature vector corresponding to a user by extracting a feature from speech data corresponding to the user;
• determining a first similarity between the feature vector and an enrolled feature vector enrolled for user verification;
• determining a second similarity between the feature vector and generalized feature vectors included in a generalized user model corresponding to generalized users;
• and
• generating a verification result of the user based on the first similarity and the second similarity,
• wherein the generating the verification result of the user comprising:
• adjusting at least one of the first similarity and a threshold based on the second similarity;
• and
• generating the verification result of the user by comparing the first similarity and the threshold,
• wherein the determining of the second similarity comprises determining the second similarity based on a Gaussian mixture model (GMM) comprising the feature vector and Gaussian distributions, and
• the Gaussian distributions correspond to a distribution of the generalized feature vectors representing the generalized users.

A user verification apparatus comprising:

• a processor configured to:
• generate a feature vector corresponding to a user based on input data input by the user;
• determine a first similarity between the feature vector and an enrolled feature vector enrolled for user verification;
• determine a second similarity between the feature vector and generalized feature vectors included in a generalized user model corresponding to generalized users;
• and
• generate a verification result of the user based on the first similarity and the second similarity,
• wherein, for the generating the verification result of the user, the processor is further configured to
• adjust at least one of the first similarity and a threshold based on the second similarity;
• and
• generate the verification result of the user by comparing the first similarity and the threshold,
• wherein the determining of the second similarity comprises determining the second similarity based on a Gaussian mixture model (GMM) comprising the feature vector and Gaussian distributions, and the Gaussian distributions correspond to a distribution of the generalized feature vectors representing the generalized users.

A speaker verification apparatus comprising:

• a processor, the processor configured to:
• generate a feature vector corresponding to a user by extracting a feature from speech data corresponding to the user;
• determine a first similarity between the feature vector and an enrolled feature vector enrolled for user verification;
• determine a second similarity between the feature vector and generalized feature vectors included in a generalized user model corresponding to generalized users;
• and
• generate a verification result of the user based on the first similarity and the second similarity,
• wherein, for the generating the verification result of the user, the processor is further configured to:
• adjust at least one of the first similarity and a threshold based on the second similarity;
• and
• generate the verification result of the user by comparing the first similarity and the threshold, and
• wherein the determining of the second similarity comprises determining the second similarity based on a Gaussian mixture model (GMM) comprising the feature vector and Gaussian distributions, and the Gaussian distributions correspond to a distribution of the generalized feature vectors representing the generalized users.

An electronic device, comprising:

• a sensor configured to receive an input from a user;
• a memory configured to store the input, an enrolled feature vector for user verification, a generalized user model corresponding to generalized users, and instructions;
• and
• a processor configured to execute the instructions to
• implement a feature extractor configured to extract a feature vector corresponding to the user from the input,
• implement a first comparator configured to determine a first parameter indicating a similarity between the feature vector and the enrolled feature vector,
• implement a second comparator configured to determine a second parameter indicating a similarity between the feature vector and generalized feature vectors included in the generalized user model, and
• verify the user based on the first parameter and the second parameter,
• wherein, for the generating the verification result of the user, the processor is further configured to:
• adjust at least one of the first similarity and a threshold based on the second similarity;
• and
• generate the verification result of the user by comparing the first similarity and the threshold, and
• wherein the determining of the second similarity comprises determining the second similarity based on a Gaussian mixture model (GMM) comprising the feature vector and Gaussian distributions, and the Gaussian distributions correspond to a distribution of the generalized feature vectors representing the generalized users.