Abstract: A method for controlling a motion tracking system, including the steps of: digitally processing sound waves detected by a plurality of microphones so as to detect a voice of a user and estimate a first direction of the user; digitally processing electromagnetic waves captured by antennas so as to detect data packets transmitted to a computing apparatus by sensors and estimate second directions of each sensor; digitally averaging the second directions so as to provide an average direction for the sensors; digitally computing a difference between the first direction and the average direction; and starting to digitally track motion of the user based on measurements of each sensor when the computed difference does not exceed a predetermined difference threshold.

Abstract: A computer-implemented process is programmed to process a source input, determine text enhancements, and present the text enhancements to apply to the sentences dictated from the source input. A text processor may use machine-learning models to process an audio input to generate sentences in a presentable format. An audio input can be processed by an automatic speech recognition model to generate electronic text. The electronic text may be used to generate sentence structures using a normalization model. A comprehension model may be used to identify instructions associated with the sentence structures and generate sentences based on the instructions and the sentence structures. An enhancement model may be used to identify enhancements to apply to the sentences. The enhancements may be presented alongside sentences generated by the comprehension model to provide the user an option to select either the enhancements or the sentences.

Abstract: Techniques for virtual assistant execution of ambiguous commands is provided. A voice instruction from a user may be received at a virtual assistant. The voice instruction may request the virtual assistant to perform a command. The command that is most likely being requested by the voice instruction from the user is identified. An ordered set of actions to execute when performing the command may be retrieved. Each action of the ordered set of actions may indicate if the action is reversible. Each action of the ordered set of actions may be executed in order until a not reversible action is reached or no further actions are in the ordered set of actions.

Abstract: The present disclosure relates to a text-to-speech synthesis method using machine learning based on a sequential prosody feature. The text-to-speech synthesis method includes receiving input text, receiving a sequential prosody feature, and generating output speech data for the input text reflecting the received sequential prosody feature by inputting the input text and the received sequential prosody feature to an artificial neural network text-to-speech synthesis model.

Abstract: A voice control interactive system and method to provide a hands-free operation for the operator to monitor and control multiple conveyors in a warehouse. The system comprises a first computing device and a second computing device. The first computing device receives an audio signal generated by a second computing device and generates a control signal and a response signal in response to the audio signal. The audio signal comprises information relating to a verbal command spoken by an operator associated with the second computing device. The response signal comprises information relating to a response for the verbal command, wherein the information is generated based on a location of the second computing device. The control signal comprises information to control a conveyor.

Abstract: Various embodiments of a system and associated method for anonymization of text without losing semantic utility of text by extracting a latent embedding representation of content with respect to a given task and by learning an optimal strategy for text embedding manipulation to satisfy both privacy and utility requirements are disclosed herein. In particular, the system balances private attribute obfuscation with retained semantic utility.

Abstract: An illustrative collaboration space provider system provides a virtual collaboration session that allows for audio communication between a user and one or more other users virtually located within a virtual collaboration space. The user is represented by an avatar located at an avatar location within the virtual collaboration space. The collaboration space provider system receives user input from the user, the user input representative of a voice origination location that is within the virtual collaboration space and is distinct from the avatar location. During the virtual collaboration session, the collaboration space provider system simulates propagation within the virtual collaboration space of a voice communication spoken by the user. The propagation of the voice communication is simulated to originate from the voice origination location and not from the avatar location. Corresponding methods and systems are also disclosed.

Abstract: Example techniques involve suppressing a wake word response to a local wake word. An example implementation involves a playback device receiving audio content for playback by the playback device and providing a sound data stream representing the received audio content to a voice assistant service (VAS) wake-word engine and a local keyword engine. The playback device plays back a first portion of the audio content and detects, via the local keyword engine, that a second portion of the received audio content includes sound data matching one or more particular local keywords. Before the second portion of the received audio content is played back, the playback device disables a local keyword response of the local keyword engine to the one or more particular local keywords and then plays back the second portion of the audio content via one or more speakers.

Abstract: In response to a user interacting with a tangible peripheral assistant control device (e.g., depressing a button of the device), causing an automated assistant to perform one or more actions. The action(s) performed can be based on input previously provided by the user in configuring the peripheral assistant control device. The action(s) performed in response to interaction with the peripheral assistant control device can vary based on one or more conditions, such as which user is currently active, where the peripheral assistant control device is currently located (which can optionally be inferred based on which of multiple assistant computing devices the button is paired with), and/or the current state of one or more smart devices and/or other devices (e.g., as determined based on a device topology). A utility of the peripheral assistant control device can be automatically extended beyond what was specifically requested by a user during configuration.

Abstract: Disclosed is an LPC residual signal encoding/decoding apparatus of an MDCT based unified voice and audio encoding device. The LPC residual signal encoding apparatus analyzes a property of an input signal, selects an encoding method of an LPC filtered signal, and encode the LPC residual signal based on one of a real filterbank, a complex filterbank, and an algebraic code excited linear prediction (ACELP).

Abstract: A method and system are provided for training a machine-learning (ML) system/module and to provide an ML model. In one embodiment, a method includes using a labeled entities set to train a machine learning (ML) system, to obtain an ML model, and using the trained ML model to predict labels for entities in an unlabeled entities set, yielding a machine-labeled entities set. One or more individual ML models may be trained and used in this way, where each individual ML model corresponds to a respective document source. The document sources can be identified via classification of a corpus of documents. The prediction of labels provides a respective confidence score for each machine-labeled entity. The method also includes selecting from the machine-labeled entities set, a subset of machine-labeled entities having a respective confidence score at least equal to a threshold confidence score; and updating the labeled entities set by adding thereto the selected subset of machine-labeled entities.

Abstract: Audio encoder for encoding a multichannel signal is shown. The audio encoder includes a downmixer for downmixing the multichannel signal to obtain a downmix signal, a linear prediction domain core encoder for encoding the downmix signal, wherein the downmix signal has a low band and a high band, wherein the linear prediction domain core encoder is configured to apply a bandwidth extension processing for parametrically encoding the high band, a filterbank for generating a spectral representation of the multichannel signal, and a joint multichannel encoder configured to process the spectral representation including the low band and the high band of the multichannel signal to generate multichannel information.

Abstract: Prevention of voice misappropriation in voice interaction/response systems. The system relies on telemetry data, including thermal data of components to determine whether a received voice command was made by actual voice. If the voice command is determined to have been made by an actual voice, a response to the command is generated and transmitted, otherwise if the voice command is determined to have likely not been made by an actual voice (e.g., artificial means replicating a voice, such as a laser or the like), no response to the command is transmitted or action taken with respect to the command.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating author vectors. One of the methods includes obtaining a set of sequences of words, the set of sequences of words comprising a plurality of first sequences of words and, for each first sequence of words, a respective second sequence of words that follows the first sequence of words, wherein each first sequence of words and each second sequence of words has been classified as being authored by a first author; and training a neural network system on the first sequences and the second sequences to determine an author vector for the first author, wherein the author vector characterizes the first author.

Abstract: Sentiment analysis is a task in natural language processing. The embodiments are directed to using a generative language model to extract an aspect term, aspect category and their corresponding polarities. The generative language model may be trained as a single, joint, and multi-task model. The single-task generative language model determines a term polarity from the aspect term in the sentence or a category polarity from an aspect category in the sentence. The joint-task generative language model determines both the aspect term and the term polarity or the aspect category and the category polarity. The multi-task generative language model determines the aspect term, term polarity, aspect category and category polarity of the sentence.

Abstract: The invention is directed towards a an audio scrubbing system that allows for scrubbing recognized voice commands from audio data and replacing the recognized voice commands with environment audio data. Specifically, as a user captures video and audio data via a HMD, audio data captured by the HMD may be processed by an audio scrubbing module to identify voice commands in the audio data that are used for controlling the HMD. When a voice command is identified in the audio data, timestamps corresponding to the voice command may be determined. Filler audio data may then be generated to imitate the environment by processing at least a portion of the audio data by a neural network of a machine learning model. The filler audio data may then be used to replace the audio data corresponding to the identified voice commands, thereby scrubbing the voice command from the audio data.

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: Support for natural language expressions is provided by the use of semantic grammars that describe the structure of expressions in that grammar and that construct the meaning of a corresponding natural language expression. A semantic grammar extension mechanism is provided, which allows one semantic grammar to be used in the place of another semantic grammar. This enriches the expressivity of semantic grammars in a simple, natural, and decoupled manner.

Abstract: Conventional audio compression technologies perform a standardized signal transformation, independent of the type of the content. Multi-channel signals are decomposed into their signal components, subsequently quantized and encoded. This is disadvantageous due to lack of knowledge on the characteristics of scene composition, especially for e.g. multi-channel audio or Higher-Order Ambisonics (HOA) content. A method for decoding an encoded bitstream of multi-channel audio data and associated metadata is provided, including transforming the first Ambisonics format of the multi-channel audio data to a second Ambisonics format representation of the multi-channel audio data, wherein the transforming maps the first Ambisonics format of the multi-channel audio data into the second Ambisonics format representation of the multi-channel audio data.

Abstract: Systems and methods for optimizing voice detection via a network microphone device are disclosed herein. In one example, individual microphones of a network microphone device detect sound. The sound data is captured in a first buffer and analyzed to detect a trigger event. Metadata associated with the sound data is captured in a second buffer and provided to at least one network device to determine at least one characteristic of the detected sound based on the metadata. The network device provides a response that includes an instruction, based on the determined characteristic, to modify at least one performance parameter of the NMD. The NMD then modifies the at least one performance parameter based on the instruction.