Abstract: Embodiments described herein provide for audio processing operations that evaluate characteristics of audio signals that are independent of the speaker's voice. A neural network architecture trains and applies discriminatory neural networks tasked with modeling and classifying speaker-independent characteristics. The task-specific models generate or extract feature vectors from input audio data based on the trained embedding extraction models. The embeddings from the task-specific models are concatenated to form a deep-phoneprint vector for the input audio signal. The DP vector is a low dimensional representation of the each of the speaker-independent characteristics of the audio signal and applied in various downstream operations.

Abstract: Embodiments described herein provide for systems and methods for implementing a neural network architecture for spoof detection in audio signals. The neural network architecture contains a layers defining embedding extractors that extract embeddings from input audio signals. Spoofprint embeddings are generated for particular system enrollees to detect attempts to spoof the enrollee's voice. Optionally, voiceprint embeddings are generated for the system enrollees to recognize the enrollee's voice. The voiceprints are extracted using features related to the enrollee's voice. The spoofprints are extracted using features related to features of how the enrollee speaks and other artifacts. The spoofprints facilitate detection of efforts to fool voice biometrics using synthesized speech (e.g., deepfakes) that spoof and emulate the enrollee's voice.

Abstract: Methods, systems, and apparatuses for audio event detection, where the determination of a type of sound data is made at the cluster level rather than at the frame level. The techniques provided are thus more robust to the local behavior of features of an audio signal or audio recording. The audio event detection is performed by using Gaussian mixture models (GMMs) to classify each cluster or by extracting an i-vector from each cluster. Each cluster may be classified based on an i-vector classification using a support vector machine or probabilistic linear discriminant analysis. The audio event detection significantly reduces potential smoothing error and avoids any dependency on accurate window-size tuning. Segmentation may be performed using a generalized likelihood ratio and a Bayesian information criterion, and the segments may be clustered using hierarchical agglomerative clustering. Audio frames may be clustered using K-means and GMMs.

Abstract: Utterances of at least two speakers in a speech signal may be distinguished and the associated speaker identified by use of diarization together with automatic speech recognition of identifying words and phrases commonly in the speech signal. The diarization process clusters turns of the conversation while recognized special form phrases and entity names identify the speakers. A trained probabilistic model deduces which entity name(s) correspond to the clusters.

Abstract: A system for generating channel-compensated features of a speech signal includes a channel noise simulator that degrades the speech signal, a feed forward convolutional neural network (CNN) that generates channel-compensated features of the degraded speech signal, and a loss function that computes a difference between the channel-compensated features and handcrafted features for the same raw speech signal. Each loss result may be used to update connection weights of the CNN until a predetermined threshold loss is satisfied, and the CNN may be used as a front-end for a deep neural network (DNN) for speaker recognition/verification. The DNN may include convolutional layers, a bottleneck features layer, multiple fully-connected layers, and an output layer. The bottleneck features may be used to update connection weights of the convolutional layers, and dropout may be applied to the convolutional layers.

Abstract: A computer may segment a noisy audio signal into audio frames and execute a deep neural network (DNN) to estimate an instantaneous function of clean speech spectrum and noisy audio spectrum in the audio frame. This instantaneous function may correspond to a ratio of an a-priori signal to noise ratio (SNR) and an a-posteriori SNR of the audio frame. The computer may add estimated instantaneous function to the original noisy audio frame to output an enhanced speech audio frame.

Abstract: Systems and methods may generate, by a computer, a voice model for an enrollee based upon a set of one or more features extracted from a first audio sample received at a first time; receive at a second time a second audio sample associated with a caller; generate a likelihood score for the second audio sample by applying the voice model associated with the enrollee on the set of features extracted from the second audio sample associated with the caller, the likelihood score indicating a likelihood that the caller is the enrollee; calibrate the likelihood score based upon a time interval from the first time to the second time and at least one of: an enrollee age at the first time and an enrollee gender; and authenticate the caller as the enrollee upon the computer determining that the likelihood score satisfies a predetermined threshold score.

Abstract: Described herein are systems and methods for improved audio analysis using a computer-executed neural network having one or more in-network data augmentation layers. The systems described herein help ease or avoid unwanted strain on computing resources by employing the data augmentation techniques within the layers of the neural network. The in-network data augmentation layers will produce various types of simulated audio data when the computer applies the neural network on an inputted audio signal during a training phase, enrollment phase, and/or testing phase. Subsequent layers of the neural network (e.g., convolutional layer, pooling layer, data augmentation layer) ingest the simulated audio data and the inputted audio signal and perform various operations.

Abstract: Utterances of at least two speakers in a speech signal may be distinguished and the associated speaker identified by use of diarization together with automatic speech recognition of identifying words and phrases commonly in the speech signal. The diarization process clusters turns of the conversation while recognized special form phrases and entity names identify the speakers. A trained probabilistic model deduces which entity name(s) correspond to the clusters.

Abstract: A system for generating channel-compensated features of a speech signal includes a channel noise simulator that degrades the speech signal, a feed forward convolutional neural network (CNN) that generates channel-compensated features of the degraded speech signal, and a loss function that computes a difference between the channel-compensated features and handcrafted features for the same raw speech signal. Each loss result may be used to update connection weights of the CNN until a predetermined threshold loss is satisfied, and the CNN may be used as a front-end for a deep neural network (DNN) for speaker recognition/verification. The DNN may include convolutional layers, a bottleneck features layer, multiple fully-connected layers and an output layer. The bottleneck features may be used to update connection weights of the convolutional layers, and dropout may be applied to the convolutional layers.

Abstract: Disclosed are systems and methods including computing-processes executing machine-learning architectures for voice biometrics, in which the machine-learning architecture implements one or more language compensation functions. Embodiments include an embedding extraction engine (sometimes referred to as an “embedding extractor”) that extracts speaker embeddings and determines a speaker similarity score for determine or verifying the likelihood that speakers in different audio signals are the same speaker. The machine-learning architecture further includes a multi-class language classifier that determines a language likelihood score that indicates the likelihood that a particular audio signal includes a spoken language. The features and functions of the machine-learning architecture described herein may implement the various language compensation techniques to provide more accurate speaker recognition results, regardless of the language spoken by the speaker.

Abstract: Disclosed are systems and methods including computing-processes executing machine-learning architectures implementing label distribution loss functions to improve age estimation performance and generalization. The machine-learning architecture includes a front-end neural network architecture defining a speaker embedding extraction engine of the machine-learning architecture, and a backend neural network architecture defining an age estimation engine of the machine-learning architecture. The embedding extractor is trained to extract low-level acoustic features of a speaker's speech, such as mel-frequency cepstral coefficients (MFCCs), from audio signals, and then extract a feature vector or speaker embedding vector that mathematically represents the low-level features of the speaker. The age estimator is trained to generate an estimated age for the speaker and a Gaussian probability distribution around the estimated age, by applying the various types of layers of the age estimator on the speaker embedding.

Abstract: The present invention is directed to a deep neural network (DNN) having a triplet network architecture, which is suitable to perform speaker recognition. In particular, the DNN includes three feed-forward neural networks, which are trained according to a batch process utilizing a cohort set of negative training samples. After each batch of training samples is processed, the DNN may be trained according to a loss function, e.g., utilizing a cosine measure of similarity between respective samples, along with positive and negative margins, to provide a robust representation of voiceprints.

Abstract: Embodiments include a computer executing voice biometric machine-learning for speaker recognition. The machine-learning architecture includes embedding extractors that extract embeddings for enrollment or for verifying inbound speakers, and embedding convertors that convert enrollment voiceprints from a first type of embedding to a second type of embedding. The embedding convertor maps the feature vector space of the first type of embedding to the feature vector space of the second type of embedding. The embedding convertor takes as input enrollment embeddings of the first type of embedding and generates as output converted enrolled embeddings that are aggregated into a converted enrolled voiceprint of the second type of embedding.

Abstract: Disclosed are systems and methods including computing-processes executing machine-learning architectures extract vectors representing disparate types of data and output predicted identities of users accessing computing services, without express identity assertions, and across multiple computing services, analyzing data from multiple modalities, for various user devices, and agnostic to architectures hosting the disparate computing service. The system invokes the identification operations of the machine-learning architecture, which extracts biometric embeddings from biometric data and context embeddings representing all or most of the types of metadata features analyzed by the system. The context embeddings help identify a subset of potentially matching identities of possible users, which limits the number of biometric-prints the system compares against an inbound biometric embedding for authentication.

Abstract: Disclosed are systems and methods including computing-processes executing machine-learning architectures extract vectors representing disparate types of data and output predicted identities of users accessing computing services, without express identity assertions, and across multiple computing services, analyzing data from multiple modalities, for various user devices, and agnostic to architectures hosting the disparate computing service. The system invokes the identification operations of the machine-learning architecture, which extracts biometric embeddings from biometric data and context embeddings representing all or most of the types of metadata features analyzed by the system. The context embeddings help identify a subset of potentially matching identities of possible users, which limits the number of biometric-prints the system compares against an inbound biometric embedding for authentication.

Abstract: An automated speaker verification (ASV) system incorporates a first deep neural network to extract deep acoustic features, such as deep CQCC features, from a received voice sample. The deep acoustic features are processed by a second deep neural network that classifies the deep acoustic features according to a determined likelihood of including a spoofing condition. A binary classifier then classifies the voice sample as being genuine or spoofed.

Abstract: The present invention is directed to a deep neural network (DNN) having a triplet network architecture, which is suitable to perform speaker recognition. In particular, the DNN includes three feed-forward neural networks, which are trained according to a batch process utilizing a cohort set of negative training samples. After each batch of training samples is processed, the DNN may be trained according to a loss function, e.g., utilizing a cosine measure of similarity between respective samples, along with positive and negative margins, to provide a robust representation of voiceprints.

Abstract: A score indicating a likelihood that a first subject is the same as a second subject may be calibrated to compensate for aging of the first subject between samples of age-sensitive biometric characteristics. Age of the first subject obtained at a first sample time and age of the second subject obtained at a second sample time may be averaged, and an age approximation may be generated based on at least the age average and an interval between the first and second samples. The age approximation, the interval between the first and second sample times, and an obtained gender of the subject are used to calibrate the likelihood score.

Abstract: A score indicating a likelihood that a first subject is the same as a second subject may be calibrated to compensate for aging of the first subject between samples of age-sensitive biometric characteristics. Age of the first subject obtained at a first sample time and age of the second subject obtained at a second sample time may be averaged, and an age approximation may be generated based on at least the age average and an interval between the first and second samples. The age approximation, the interval between the first and second sample times, and an obtained gender of the subject are used to calibrate the likelihood score.