Presented are a combined learning method and device using a transformed loss function and feature enhancement based on a deep neural network for speaker recognition that is robust to a noisy environment. The combined learning method using the transformed loss function and the feature enhancement based on the deep neural network for speaker recognition that is robust to the noisy environment, according to an embodiment, may comprise: a preprocessing step for learning to receive, as an input, a speech signal and remove a noise or reverberation component by using at least one of a beamforming algorithm and a dereverberation algorithm using the deep neural network; a speaker embedding step for learning to classify an utterer from the speech signal, from which a noise or reverberation component has been removed, by using a speaker embedding model based on the deep neural network; and a step for, after connecting a deep neural network model included in at least one of the beamforming algorithm and the dereverberation algorithm and the speaker embedding model, for speaker embedding, based on the deep neural network, performing combined learning by using a loss function.

Presented are a combined learning method and device using a transformed loss function and feature enhancement based on a deep neural network for speaker recognition that is robust to a noisy environment. The combined learning method using the transformed loss function and the feature enhancement based on the deep neural network for speaker recognition that is robust to the noisy environment, according to an embodiment, may comprise: a preprocessing step for learning to receive, as an input, a speech signal and remove a noise or reverberation component by using at least one of a beamforming algorithm and a dereverberation algorithm using the deep neural network; a speaker embedding step for learning to classify an utterer from the speech signal, from which a noise or reverberation component has been removed, by using a speaker embedding model based on the deep neural network; and a step for, after connecting a deep neural network model included in at least one of the beamforming algorithm and the dereverberation algorithm and the speaker embedding model, for speaker embedding, based on the deep neural network, performing combined learning by using a loss function.

A speaker recognition method and apparatus receives a first voice signal of a speaker, generates a second voice signal by enhancing the first voice signal through speech enhancement, generates a multi-channel voice signal by associating the first voice signal with the second voice signal, and recognizes the speaker based on the multi-channel voice signal.

To generate substantially condition-invariant and speaker-discriminative features, embodiments are associated with a feature extractor capable of extracting features from speech frames based on first parameters, a speaker classifier capable of identifying a speaker based on the features and on second parameters, and a condition classifier capable of identifying a noise condition based on the features and on third parameters. The first parameters of the feature extractor and the second parameters of the speaker classifier are trained to minimize a speaker classification loss, the first parameters of the feature extractor are further trained to maximize a condition classification loss, and the third parameters of the condition classifier are trained to minimize the condition classification loss.

Provided are a method and device for recognizing a speaker by using a resonator. The method of recognizing the speaker includes receiving a plurality of electrical signals corresponding to a speech of the speaker from a plurality of resonators having different resonance bands; obtaining a difference of magnitudes of the plurality of electrical signals; and recognizing the speaker based on the difference of magnitudes of the plurality of electrical signals.

Provided are a method and device for recognizing a speaker by using a resonator. The method of recognizing the speaker includes receiving a plurality of electrical signals corresponding to a speech of the speaker from a plurality of resonators having different resonance bands; obtaining a difference of magnitudes of the plurality of electrical signals; and recognizing the speaker based on the difference of magnitudes of the plurality of electrical signals.

Techniques disclosed herein enable training and/or utilizing speaker dependent (SD) speech models which are personalizable to any user of a client device. Various implementations include personalizing a SD speech model for a target user by processing, using the SD speech model, a speaker embedding corresponding to the target user along with an instance of audio data. The SD speech model can be personalized for an additional target user by processing, using the SD speech model, an additional speaker embedding, corresponding to the additional target user, along with another instance of audio data. Additional or alternative implementations include training the SD speech model based on a speaker independent speech model using teacher student learning.

Techniques disclosed herein enable training and/or utilizing speaker dependent (SD) speech models which are personalizable to any user of a client device. Various implementations include personalizing a SD speech model for a target user by processing, using the SD speech model, a speaker embedding corresponding to the target user along with an instance of audio data. The SD speech model can be personalized for an additional target user by processing, using the SD speech model, an additional speaker embedding, corresponding to the additional target user, along with another instance of audio data. Additional or alternative implementations include training the SD speech model based on a speaker independent speech model using teacher student learning.

There is provided a method that includes (a) obtaining a first voice vector that was derived from a signal of a voice that was sampled at a first sampling frequency, (b) obtaining a second voice vector that was derived from a signal of a voice that was sampled at a second sampling frequency, (c) mapping the second voice vector into a mapped voice vector in accordance with a machine learning model, and (d) comparing the first voice vector to the mapped voice vector to yield a score that indicates a probability that the first voice vector and the second voice vector originated from a same person.

This application describes methods and apparatus for speaker recognition. An apparatus according to an embodiment has an analyzer for analyzing each frame of a sequence of frames of audio data which correspond to speech sounds uttered by a user to determine at least one characteristic of the speech sound of that frame. An assessment module determines, for each frame of audio data, a contribution indicator of the extent to which that frame of audio data should be used for speaker recognition processing based on the determined characteristic of the speech sound. Said contribution indicator comprises a weighting to be applied to each frame in the speaker recognition processing. In this way frames which correspond to speech sounds that are of most use for speaker discrimination may be emphasized and/or frames which correspond to speech sounds that are of least use for speaker discrimination may be de-emphasized.