Abstract: A method includes receiving a reference audio signal corresponding to reference speech spoken by a target speaker with atypical speech, and generating, by a speaker embedding network configured to receive the reference audio signal as input, a speaker embedding for the target speaker. The speaker embedding conveys speaker characteristics of the target speaker. The method also includes receiving a speech conversion request that includes input audio data corresponding to an utterance spoken by the target speaker associated with the atypical speech. The method also includes biasing, using the speaker embedding generated for the target speaker by the speaker embedding network, a speech conversion model to convert the input audio data corresponding to the utterance spoken by the target speaker associated with atypical speech into an output canonical representation of the utterance spoken by the target speaker.

Abstract: A method for remote attestation includes establishing, using a cryptographic protocol, a communication session between a first computing device and a second computing device. The communication session includes communications encrypted by an ephemeral session key. The method includes receiving, at the first communication device via the communication session, from the second computing device, an attestation request requesting the first computing device to provide an attestation report. The method includes generating, by the first computing device, the attestation report based on the ephemeral session key and sending, using the communication session, the attestation report to the second computing device.

Abstract: A method includes generating a base model by training with a first dataset of data pairs and generating an adapted model by training the base model on a second dataset of data pairs. The method also includes determining a contrastive score for each data pair of a third dataset of data pairs using the base model and the adapted model. The contrastive score is indicative of a probability of quality of the respective data pair. The method also includes training a target model using the data pairs of the third dataset and the contrastive scores.

Abstract: The present disclosure provides systems and methods that makes use of one or more image sensors of a device to provide users with information relating to nearby points of interest. The image sensors may be used to detect features and/or objects in the field of view of the image sensors. Pose data, including a location and orientation of the device is then determined based on the one or more detected features and/or objects. A plurality of points of interest that are within a geographical area that is dependent on the pose data are then determined. The determination may, for instance, be made by querying a mapping database for points of interest that are known to be located within a particular distance of the location of the user. The device then provides information to the user indicating one or more of the plurality of points of interest.

Abstract: The present document describes a camera device with an adjustable stand. The camera device includes a head assembly and a stand assembly pivotally connected together by a stem forming a hinge. The stem provides a 360-degree range of pan and a 45-degree range of tilt of the head assembly relative to the stand assembly. The stand assembly is rotatably movable relative to the head assembly to configure the camera device in different configuration states, including a tabletop state and a wall state. The tabletop state has a low profile for resting on a horizontal surface, and the wall state has a high profile, which provides additional clearance between the head assembly and a vertical surface to which the stand assembly is affixed. In the wall state, a cable of the camera device can be routed through and constrained by the stand assembly.

Abstract: A method of training a speech model includes receiving, at a voice-enabled device, a fixed set of training utterances where each training utterance in the fixed set of training utterances includes a transcription paired with a speech representation of the corresponding training utterance. The method also includes sampling noisy audio data from an environment of the voice-enabled device. For each training utterance in the fixed set of training utterances, the method further includes augmenting, using the noisy audio data sampled from the environment of the voice-enabled device, the speech representation of the corresponding training utterance to generate noisy audio samples and pairing each of the noisy audio samples with the corresponding transcription of the corresponding training utterance. The method additionally includes training a speech model on the noisy audio samples generated for each speech representation in the fixed set of training utterances.

Abstract: Various arrangements are detailed herein related to managing video recording. A system can include a network-enabled video camera that captures a video stream of a location and streams the video stream to a cloud-based home assistant integration platform. The system can include a home assistant device comprising a microphone and wireless network interface, the home assistant device configured to receive, via the microphone, a spoken command that requests video, captured by the network-enabled video camera, be recorded. The cloud-based home assistant integration platform may be configured to analyze the video stream captured using the network-enabled video camera to determine an identity of the user. The platform may be further configured to store a portion of the video stream linked to a user account of the user based on the spoken command and the identity of the user.

Abstract: Methods, and systems, including computer programs encoded on computer storage media for generating machine learning models. A method includes receiving, through a GUI, user selection of a mode button displayed in the GUI, wherein the mode button, when selected, causes the GUI to display a first set of user-selectable buttons that correspond to respective machine learning routines, and when not selected, causes the GUI to display a second set of user-selectable buttons that correspond to respective machine learning sub-routines; in response to receiving user selection of the mode button, displaying the first set of user-selectable buttons; receiving user selection of one or more of the first set of user-selectable buttons; displaying a graphical representation of a machine learning model defined by machine learning routines corresponding to the user selected one or more of the first set of user-selectable buttons and generating a file representing the machine learning model.

Abstract: A method of encoding and decoding user interface interactions includes receiving a query from a user captured by an assistant-enabled device associated with the user. The query includes a user intent for interacting with an application. The method includes obtaining, for the application, a set of categorized actions. Each categorized action of the set of categorized actions is associated with one or more parameters and represents a high-level description of the user intent of the user interacting with the application. For each respective categorized action of the set of categorized actions, the method includes selecting a respective script associated with the respective categorized action that performs one or more low-level interactions with the application and executing the respective script to represent the user intent for interacting with the application.

Abstract: This document describes systems and techniques directed at compensating for non-uniform luminance in curved-edge displays. In aspects, a computing device having a curved-edge display and a luminance manager is configured to receive an indication of a luminance that is, or is intended to be, displayed by pixels of the curved-edge display. Responsive to and based on the received indication of the luminance and a non-uniform luminance, the luminance manager determines a luminance modification for the pixels of the curved-edge display. Based on the determined luminance modification, the luminance manager modifies the luminance that is displayed or modifies the intended luminance that is intended to be displayed by pixels of the curved-edge display effective to compensate for the non-uniform luminance.

Abstract: A contactless health monitoring device may perform a beam steering process that creates a plurality of beam-steered radar data streams from the received radar data stream. The contactless health monitoring device may determine breathing displacement for a user in relation to time for each spatial zone radar data stream. The contactless health monitoring device may analyze the breathing displacement for the user in relation to time for each spatial zone radar data stream. The contactless health monitoring device may output a screening result based on analyzing the breathing displacement for the user.

Abstract: Techniques and apparatuses are described for radio access technology identifiers. In aspects, a base station receives air interface resource prospective usage information associated with communications over a first wireless link of at least two wireless links that use different Radio Access Technologies (RATS). The base station then allocates the sharable air interface resource between the at least two wireless links by analyzing the air interface resource prospective usage information. In aspects, the base station transmits a RAT-identifier-presence indicator that communicates a presence of a RAT identifier by indicating one or more downlink air interface resources used to transmit the RAT identifier, where the RAT identifier indicates the allocation of the sharable air interface resource between the at least two wireless links. The base station then transmits the RAT identifier using the one or more downlink air interface resources indicated by the RAT-identifier-presence indicator.

Abstract: While an assistant-enabled device is playing back media content, a method includes receiving a contextual signal from an environment of the assistant-enabled device and executing an event recognition routine to determine whether the received contextual signal is indicative of an event that conflicts with the playback of the media content from the assistant-enabled device. When the event recognition routine determines that the received contextual signal is indicative of the event that conflicts with the playback of the media content, the method also includes adjusting content playback settings of the assistant-enabled device.

Abstract: Techniques and apparatuses are described that implement an ultrasonic sensor capable of detecting user presence. This ultrasonic sensor can detect user presence without relying on time-of-flight techniques. In particular, the ultrasonic sensor can determine that a user is present based on the occlusion of at least one receiving transducer (e.g., microphone occlusion), the occlusion of at least one transmitting transducer (e.g., speaker occlusion), or a detected change in an audible noise floor of at least one transducer. In this way, the ultrasonic sensor can continue to detect user presence in situations in which a user occludes one or more transducers of the ultrasonic sensor. The ultrasonic sensor can also control operation of another component within a computing device based on the presence of the user to improve the user experience and/or improve power management.

Abstract: A method for training a model comprises obtaining a set of labeled training samples each associated with a given label. For each labeled training sample, the method includes generating a pseudo label and estimating a weight of the labeled training sample indicative of an accuracy of the given label. The method also includes determining whether the weight of the labeled training sample satisfies a weight threshold. When the weight of the labeled training sample satisfies the weight threshold, the method includes adding the labeled training sample to a set of cleanly labeled training samples. Otherwise, the method includes adding the labeled training sample to a set of mislabeled training samples. The method includes training the model with the set of cleanly labeled training samples using corresponding given labels and the set of mislabeled training samples using corresponding pseudo labels.

Abstract: This document describes methods, devices, systems, and means for a switching scheme for opting in and out of multi-user orthogonal frequency-division multiple access (MU-OFDMA). In one aspect, an electronic device enters the MU-OFDMA mode to communicate via a wireless network over a shared-channel bandwidth. During the MU-OFDMA mode, the electronic device determines that an uplink-queue size is greater than a first threshold size. Responsive to the determining, the electronic device opts out of the MU-OFDMA mode and enters a single-user mode to contend for a transmit channel for transmitting uplink data.

Abstract: Techniques and devices for securing return communication through application uniform resource locators are described for commissioning a joiner device to a home area network by an initiator device in which the initiator device obtains a Responder Access Uniform Resource Locator (URL) and using the obtained Responder Access URL, generates an Augmented Responder Access URL. The initiator device accesses the Augmented Responder Access URL at a responder, which causes the responder to generate a Responder Payload. The initiator device accesses an Augmented Initiator Response URL including the generated Responder Payload and recovers the Responder Payload, the recovery of the Responder Payload causing the initiator device to commission the joiner device to the home area network.

Abstract: A computer-implemented method when executed by data processing hardware of a user device causes the data processing hardware to perform operations. The operations include obtaining, from a message server, an encrypted message encrypted by a single-use data encryption key (DEK) and an encrypted DEK including the single-use DEK encrypted by a public key (PK). The operations also include transmitting, to a key access control list server (KACLS), a decryption request requesting the KACLS decrypt the encrypted DEK with a PRK associated with the PK. The decryption request includes the encrypted DEK. The KACLS is independent from the message server. The operations also include receiving, from the KACLS, the single-use DEK and decrypting, using the single-use DEK, the encrypted message.

Abstract: Systems and methods for displaying media files on a device are provided. The device provides a first mode that displays a subset of media files, determined by a user query, from a plurality of media files. Once activated, the first mode persists on the device prior to becoming inactive. A second mode of the device displays the plurality of media files. The second mode is active when the first mode is inactive. An instruction is received from a user. When the instruction specifies the first mode by inclusion of a user query, a corresponding subset of files is obtained from a remote system. The subset of files is sequentially displayed until the period of time has elapsed according to the first mode. When the instruction does not specify a query, the plurality of files is polled for from a remote device and sequentially displayed according to the second mode.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for storing and accessing data in a cloud system. One of the methods includes receiving an identification of log data that records occurrences of events; receiving a specification of a plurality of different event types to be indexed; indexing the log data according to the specification and group identifiers; receiving a query specifying a reference parameter and requesting one or more predicted events; searching the indexed groups to identify a plurality of groups having events associated with the reference parameter; computing one or more predicted events, from the identified plurality of groups, that are most likely to co-occur in the indexed groups with events associated with the reference parameter; and providing the computed one or more predicted events.

Abstract: Aspects of the subject technology relate to a method for using a voice command for multiple computing devices. First voice input data is received from a first computing device associated with a user account, where the first voice input data comprises a first voice command captured at the first computing device. Second voice input data is received from a second computing device associated with the user account where the second voice input data comprises a second voice command captured at the second computing device. An intended voice command is determined based on the obtained first and second voice input data. Based on the intended voice command, a first target computing device is determined. First instructions associated with the intended voice command are provided to the first target computing device for execution.

Abstract: This document describes a light-sealing gasket with crossbar force distribution. The gasket can be used in an electronic device with a sensor package that is capable of transmitting and receiving signals and is positioned behind a display of the device. The gasket provides a shield between the receive signals and the transmit signals, prevents signal crosstalk, and protects the delicate panel layer of the display. Use of this gasket in an electronic device allows manufacturers to add more features to the device and enrich the user experience.

Abstract: A RNN-T model includes a prediction network configured to, at each of a plurality of times steps subsequent to an initial time step, receive a sequence of non-blank symbols. For each non-blank symbol the prediction network is also configured to generate, using a shared embedding matrix, an embedding of the corresponding non-blank symbol, assign a respective position vector to the corresponding non-blank symbol, and weight the embedding proportional to a similarity between the embedding and the respective position vector. The prediction network is also configured to generate a single embedding vector at the corresponding time step. The RNN-T model also includes a joint network configured to, at each of the plurality of time steps subsequent to the initial time step, receive the single embedding vector generated as output from the prediction network at the corresponding time step and generate a probability distribution over possible speech recognition hypotheses.

Abstract: Techniques and apparatuses are described for adapting an end-to-end, E2E, machine-learning, ML, configuration for processing communications transferred through an E2E communication. A network entity directs a user equipment (UE) and a base station participating in the E2E communication to implement the E2E communication by forming at least a portion of an E2E deep neural network, DNN, based on a first E2E ML configuration. The network entity determines to update the first E2E ML configuration based on a change in a participation mode of an edge compute server (ECS) in the E2E communication. The network entity identifies a second E2E ML configuration based on the change in participation mode and directs the UE or the base station to update the portion of the E2E DNN using the second E2E ML configuration.

Abstract: An ASR model includes a first encoder configured to receive a sequence of acoustic frames and generate a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The ASR model also includes a second encoder configured to receive the first higher order feature representation generated by the first encoder at each of the plurality of output steps and generate a second higher order feature representation for a corresponding first higher order feature frame. The ASR model also includes a decoder configured to receive the second higher order feature representation generated by the second encoder at each of the plurality of output steps and generate a first probability distribution over possible speech recognition hypothesis. The ASR model also includes a language model configured to receive the first probability distribution over possible speech hypothesis and generate a rescored probability distribution.

Abstract: The present document describes techniques associated with a blind battery connector. The blind battery connector described herein enables a user to blindly engage, safely and securely, a battery connector with a system-side connector. In aspects, the blind battery connector includes polarity-oriented magnets at both the battery connector and the system-side connector to automatically align and engage the battery connector with the system-side connector with correct orientation. The magnets may be embedded or removably assembled to the battery connector and the system-side connector. The blind battery connector controls initial alignment of the battery connector for coupling with the system-side connector and provides additional mechanical strength to the coupling against drop, vibration, and shock.

Abstract: A method includes executing a pool of primary virtual machine (VM) instances, each primary VM instance executing a corresponding individual service instance, and instantiating a shared secondary VM instance. The method includes identifying unavailability of a particular primary VM instance of the pool of primary VM instances, and causing the corresponding individual service instance executing on the particular primary VM instance to failover to the shared secondary VM instance to commence executing the corresponding individual service instance. The method includes, after the failover to the shared secondary VM instance, determining a difference between a current resource level of the shared secondary VM instance and a target resource level associated with the corresponding individual service instance, and adjusting the current resource level of the secondary VM instance based on the difference.

Abstract: Techniques and apparatuses are described that enable memory request timeouts using a common counter. A memory request is received, and a common count timeout is generated for the memory request based on a common count at a time of receipt and a latency requirement of the memory request. Common count timeouts of one or more related memory requests within a memory request buffer (if they exist) are adjusted as needed, and the memory request is placed in the memory request buffer. The common count is incremented, and the memory request is indicated as timed out in response to an incrementation of the common count matching the common count timeout for the memory request.

Abstract: This application is directed to a doorbell camera for illuminating and capturing scenes. The doorbell camera includes at least a subset of processors for operating a camera module, an image sensor having a field of view of a scene and configured to capture video of a portion of the scene, one or more infrared (IR) illuminators for providing illumination, a waterproof button assembly, and a microphone and a speaker for enabling a real-time conversation between a visitor located at the doorbell camera and a user of a remote client device. The waterproof button assembly is configured to receive a user press on a button top, block water from entering the electronic device, and display a visual pattern uniformly at a peripheral region of the button assembly using LEDs and light guide component that are disposed under the button top.

Abstract: Implementations relate to providing information items for display during a communication session. In some implementations, a computer-implemented method includes receiving, during a communication session between a first computing device and a second computing device, first media content from the communication session. The method further includes determining a first information item for display in the communication session based at least in part on the first media content. The method further includes sending a first command to at least one of the first computing device and the second computing device to display the first information item.

Abstract: A memory device includes at least one memory cell array block and a control logic. The memory cell array block includes multiple layers of memory cells and word line layers provided corresponding to individual layers of memory cells. The memory cell array block is divided into at least two memory cell array subblocks, each subblock comprising a number of layers of memory cells and word line layers provided corresponding to individual layers of memory cells. The control logic is coupled to the memory cell array block, and configured to: erase, read or program the memory cell array block using a block mode or a subblock mode, and when the memory cell array block is erased, read, or programmed under the subblock mode, determine, at least based on a state of one of the two memory cell array subblocks, an operation strategy of the other memory cell array subblock.

Abstract: The present document describes a textile-assembly toolkit for reversible assembly of a textile to an electronic-speaker device. The toolkit includes multiple attachment features, including rigid features with matched purposefully-designed knit types that can be combined to enable repeatable, mass-producible, reversible assembly of the textile to the electronic-speaker device. The techniques described herein enable accurate alignment of the textile on the electronic-speaker device without distorting the textile's cosmetic pattern and in a manner that results in no visible edges of the textile or visible attachment features on the exterior of the electronic-speaker device. Also, the textile-assembly toolkit includes attachment features that secure the textile with sufficient tension to avoid acoustic distortion such as rub and buzz.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for detecting hotwords using a server. One of the methods includes receiving an audio signal encoding one or more utterances including a first utterance; determining whether at least a portion of the first utterance satisfies a first threshold of being at least a portion of a key phrase; in response to determining that at least the portion of the first utterance satisfies the first threshold of being at least a portion of a key phrase, sending the audio signal to a server system that determines whether the first utterance satisfies a second threshold of being the key phrase, the second threshold being more restrictive than the first threshold; and receiving tagged text data representing the one or more utterances encoded in the audio signal when the server system determines that the first utterance satisfies the second threshold.

Abstract: A joint segmenting and ASR model includes an encoder and decoder. The encoder configured to: receive a sequence of acoustic frames characterizing one or more utterances; and generate, at each output step, a higher order feature representation for a corresponding acoustic frame. The decoder configured to: receive the higher order feature representation and generate, at each output step: a probability distribution over possible speech recognition hypotheses, and an indication of whether the corresponding output step corresponds to an end of speech segment. The j oint segmenting and ASR model trained on a set of training samples, each training sample including: audio data characterizing a spoken utterance; and a corresponding transcription of the spoken utterance, the corresponding transcription having an end of speech segment ground truth token inserted into the corresponding transcription automatically based on a set of heuristic-based rules and exceptions applied to the training sample.

Abstract: A method includes receives data representing a changed chunk of data in a current revision of a data volume, the changed chunk includes data having changes from previous data of a previous revision of the data volume. The method creates a block of data representing the changed chunk of data on the object store, the object store also stores previous revision data of the previous revision. The method determines a previous index stored on the object store corresponding to the previous revision, which includes entries including at least one corresponding to the previous revision data. The method creates a revised index that updates the corresponding entry with updated entry data representing the changed chunk of data. The method includes deleting, from the object store, each particular block of data stored on the object store that is no longer associated with an entry on any index stored on the object store.

Abstract: The present disclosure describes various aspects of secure provisioning with hardware verification. In some aspects, sensitive data are provisioned to an integrated circuit (IC) device through a provisioning process. Provisioning data for the IC device are divided into a plurality of fragments, and each fragment is encrypted in one of a plurality of cryptographic keys. Corresponding cryptographic keys are generated at the IC device. The encrypted fragments are transferred to the IC device in respective secure transfer operations, each including sending a seed value to the IC device, validating integrity data configured to characterize integrated circuitry within a portion of the IC device specified by the seed value, and transferring the encrypted fragment to the IC device in response to validating the integrity data. In response to completing the secure transfer operation, the IC device may reconstruct the provisioning data from the encrypted fragments and corresponding cryptographic keys.

Abstract: Methods, systems and media for presenting media content that was advertised on a second screen device using a primary screen device are provided.

Abstract: A method for data storage includes specifying a plurality of File Systems (FSs) for use by multiple clients, including assigning to the FSs both respective global identifiers and respective client-specific names. The plurality of FSs is managed using the global identifiers, and files are stored for the clients in the FSs using the client-specific names.

Abstract: Systems, methods, and computer readable storage media that may be used to train a model based on merged common features of two or more different data types. One method includes receiving a plurality of first data elements of a first data type and a plurality of second data elements of a second data type, identifying first features of each of the plurality of first data elements, identifying second features of each of the plurality of second data elements, generating merged features by combining a first feature of the first features of each of the plurality of first data elements with a second feature of the second features of one of the plurality of second data elements, wherein the first feature and the second feature each represent a common feature, and training a model based on the merged features and at least a portion of the first features and the second features.

Abstract: A method for generating frame values using a key frame network includes receiving a text utterance having at least one phoneme, and for each respective phoneme of the at least one phoneme, predicting, using a predictive model, a fixed quantity of key frames. Each respective key frame of the fixed quantity of key frames includes a representation of a component of the respective phoneme. The method also includes generating, using the fixed quantity of key frames, a plurality of frame values. Here, each respective frame value of the plurality of frame values is representative of a fixed-duration of audio.

Abstract: Techniques and devices for expressing multicast groups using Weave traits are described for generating a multicast address in which a wireless network device receives an application group identifier and hashes the application group identifier to generate a group identifier. The wireless network device concatenates a global identifier and a Weave fabric identifier, inserts the concatenated global identifier and the Weave fabric identifier into a network prefix field of a destination address, and inserts the group identifier into a group identifier field of the destination address. The wireless network device sets the value of a plen field of the destination address to a value that indicates a length of the Weave fabric identifier and inserts the destination address into an Internet Protocol version 6 data packet.

Abstract: A method includes receiving, as input to a speech recognition model, audio data corresponding to a spoken utterance. The method also includes performing, using the speech recognition model, speech recognition on the audio data by, at each of a plurality of time steps, encoding, using an audio encoder, the audio data corresponding to the spoken utterance into a corresponding audio encoding, and decoding, using a speech recognition joint network, the corresponding audio encoding into a probability distribution over possible output labels. At each of the plurality of time steps, the method also includes determining, using an intended query (IQ) joint network configured to receive a label history representation associated with a sequence of non-blank symbols output by a final softmax layer, an intended query decision indicating whether or not the spoken utterance includes a query intended for a digital assistant.

Abstract: This application is directed to a battery having a heating element. A connector of the battery includes a first terminal, a second terminal, and a heater terminal. One or more rechargeable battery cells are electrically coupled to the first and second terminals of the connector. The heating element is in contact with a subset of the battery cells, and includes a resistive heater path that is electrically coupled to the first and heater terminals of the connector and generates heat to warm the battery when a heater voltage is applied to the heater terminal. A waterproof material is wrapped around an exterior of the heating element and battery cells and prevents ambient water from contacting the heater element and battery cells. The waterproof material includes an opening to allow at least the first and second terminals of the connector to be electrically coupled to a logic board.

Abstract: A method includes inserting a set of canary text samples into a corpus of training text samples and training an external language model on the corpus of training text samples and the set of canary text samples inserted into the corpus of training text samples. For each canary text sample, the method also includes generating a corresponding synthetic speech utterance and generating an initial transcription for the corresponding synthetic speech utterance. The method also includes rescoring the initial transcription generated for each corresponding synthetic speech utterance using the external language model. The method also includes determining a word error rate (WER) of the external language model based on the rescored initial transcriptions and the canary text samples and detecting memorization of the canary text samples by the external language model based on the WER of the external language model.

Abstract: A method for contextual biasing for speech recognition includes obtaining a base automatic speech recognition (ASR) model trained on non-biased data and a sub-model trained on biased data representative of a particular domain. The method includes receiving a speech recognition request including audio data characterizing an utterance captured in streaming audio. The method further includes determining whether the speech recognition request includes a contextual indicator indicating the particular domain. When the speech recognition request does not include the contextual indicator, the method includes generating, using the base ASR model, a first speech recognition result of the utterance by processing the audio data.

Abstract: A method includes receiving, from a speech recognizer, multiple existing candidate hypotheses for an utterance. Each existing candidate hypothesis has a corresponding likelihood score assigned by the speech recognizer. The method also includes generating, using a correction module configured to receive the multiple candidate hypotheses as input, a new candidate hypothesis and determining, using a comparison model configured to receive the corresponding likelihood score assigned to one of the multiple existing candidate hypotheses as input, a corresponding likelihood score. The method also includes ranking the multiple existing candidate hypotheses and the new candidate hypothesis based on the corresponding likelihood scores assigned to the multiple existing candidate hypothesis by the speech recognizer and the corresponding likelihood score for the new candidate hypothesis.

Abstract: A method includes receiving audio data corresponding to an utterance spoken by a user and processing, using a first recognition model, the audio data to generate a non-contextual candidate hypothesis as output from the first recognition model. The non-contextual candidate hypothesis has a corresponding likelihood score assigned by the first recognition model. The method also includes generating, using a second recognition model configured to receive personal context information, a contextual candidate hypothesis that includes a personal named entity. The method also includes scoring, based on the personal context information and the corresponding likelihood score assigned to the non-contextual candidate hypothesis, the contextual candidate hypothesis relative to the non-contextual candidate hypotheses.

Abstract: Various arrangements for monitoring for contactless human interactions and health using a single radar transmission modulation mode are provided. Radar chirps may be output by a radar sensor operating in a burst mode. The burst mode radar data stream may be monitored for a contactless human interaction performed by an active user. The burst mode radar data stream may be converted to a virtual continuous mode radar data stream. Health monitoring of an inactive user may be performed using the virtual continuous mode radar data stream.

Abstract: An automated speech recognition (ASR) model includes a first encoder, a first encoder, a second encoder, and a second decoder. The first encoder receives, as input, a sequence of acoustic frames, and generates, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The first decoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a first probability distribution over possible speech recognition hypotheses. The second encoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a second higher order feature representation for a corresponding first higher order feature frame. The second decoder receives, as input, the second higher order feature representation generated by the second encoder, and generates a second probability distribution over possible speech recognition hypotheses.

Abstract: A method includes obtaining a batch of training samples. For each particular training sample in the batch of training samples, the method includes generating, using a data value estimator model and the particular training sample, a corresponding predicted value of the particular training sample when used to train a machine learning model. The method includes selecting, based on the corresponding predicted values, a subset of the batch of training samples. For each particular training sample in the subset of the batch of training samples, the method includes determining, using the machine learning model and the particular training sample, a corresponding prediction performance measurement. The method includes adjusting one or more estimator parameter values of the data value estimator model based on the corresponding prediction performance measurements.