Abstract: Systems and methods are related to tracking an attention of a user with respect to content presented on a virtual screen, detecting a defocus event associated with a first region of the content, and determining a next focus event associated with a second region of the content. The determination can be based at least in part on the defocus event and on the tracked attention of the user. The systems and methods can include generating, based on the determined next focus event, a marker for differentiating the second region of the content from a remainder of the content, and in response to detecting a refocus event associated with the virtual screen, triggering execution of the marker associated with the second region of the content.

Abstract: Generating edge-depth values for an object, utilizing the edge-depth values in generating a 3D point cloud for the object, and utilizing the generated 3D point cloud for generating a 3D bounding shape (e.g., 3D bounding box) for the object. Edge-depth values for an object are depth values that are determined from frame(s) of vision data (e.g., left/right images) that captures the object, and that are determined to correspond to an edge of the object (an edge from the perspective of frame(s) of vision data). Techniques that utilize edge-depth values for an object (exclusively, or in combination with other depth values for the object) in generating 3D bounding shapes can enable accurate 3D bounding shapes to be generated for partially or fully transparent objects. Such increased accuracy 3D bounding shapes directly improve performance of a robot that utilizes the 3D bounding shapes in performing various tasks.

Abstract: A system including one or more processors configured to receive a transpose instruction indicating to transpose a source matrix to a result matrix, provide data elements of the source matrix to input switching circuits, reorder the data elements using the input switching circuits, provide the data elements from the input switching circuits to one or more lanes of a datapath, provide the data elements from the datapath to output switching circuits, undo the reordering of the data elements using the output switching circuits, and provide the data elements from the output switching circuits to a result matrix. Each respective lane of the datapath receiving data elements receives multiple data elements directed to different respective non-overlapping portions of the lane.

Abstract: A method includes receiving follow-on audio data captured by an assistant-enabled device, the follow-on audio data corresponding to a follow-on query spoken by a user of the assistant-enabled device to a digital assistant subsequent to the user submitting a previous query to the digital assistant. The method also includes processing, using a speech recognizer, the follow-on audio data to generate multiple candidate hypotheses, each candidate hypothesis corresponding to a candidate transcription for the follow-on query and represented by a respective sequence of hypothesized terms. For each corresponding candidate hypothesis among the multiple candidate hypotheses, the method also includes determining a corresponding similarity metric between the previous query and the corresponding candidate hypothesis and determining a transcription of the follow-on query spoken by the user based on the similarity metrics determined for the multiple candidate hypotheses.

Abstract: A circuit for performing neural network computations for a neural network is described. The circuit includes plurality of neural network layers each including a crossbar arrays. The plurality of crossbar arrays are formed in a common substrate in a stacked configuration. Each crossbar array includes a set of crosspoint devices. A respective electrical property of each of the crosspoint devices is adjustable to represent a weight value that is stored for each respective crosspoint device. A processing unit is configured to adjust the respective electrical properties of each of the crosspoint devices by pre-loading each of the crosspoint devices with a tuning signal. A value of the turning signal for each crosspoint device is a function of the weight value represented by each respective crosspoint device.

Abstract: Role switch between wirelessly paired master/slave devices may be performed without perceived glitches in audio. The devices negotiate an anchor point, such as a point in time or a point in relation to other events, for performing the role switch. To prepare for the role switch prior to the anchor point, the devices communicate a variety of information, such as information for communicating with the host device after the role switch and information for bit processing after the role switch. The slave device may use such information to serve in the master role, without the host knowing that a role switch occurred.

Abstract: A method for providing personalized responses to textual prompts using a large scale, privacy preserving, large language model (LLM) includes receiving a textual prompt from a user specifying a task for an LLM to perform, and obtaining a set of user features associated with the user. The method also includes determining, using the set of user features associated with the user, a user prompt embedding for the user, and processing, using the LLM, the textual prompt conditioned on the user prompt embedding for the user to generate a personalized response to the textual prompt. The method further includes providing the personalized response to the textual prompt for output from a user device associated with the user.

Abstract: A multimedia communication system and computer-implemented method for transmitting auxiliary display content to an end-user communication device to be rendered on a display device with a special effect to emphasize an image included in the auxiliary display content, comprising analyzing image data included in an auxiliary display content to detect an object image or a background image, determining a special effect based on the analysis of the image data, applying the special effect to the auxiliary display content to modify display properties for the auxiliary display content such that the object image is emphasized or pops out, and sending the auxiliary display content with modified display properties to the end-user communication device. The special effect can comprise a non-customization special effect, a simple foreground special effect or a selective foreground special effect.

Abstract: The present disclosure provides a mechanism to synchronously drive distributed speakers around a user based on localization outputs of ultra wideband (UWB) communication chips already existing in devices. Distances may be determined between a user device, such as a phone or wearable, and a plurality of distributed speakers or other devices. Based on an intersection point of such distances, the user's location can be identified. Such location can be used to modify how audio is played on each of the plurality of distributed speakers.

Abstract: A method includes receiving an input text sequence to be synthesized into speech in a first language and obtaining a speaker embedding, the speaker embedding specifying specific voice characteristics of a target speaker for synthesizing the input text sequence into speech that clones a voice of the target speaker. The target speaker includes a native speaker of a second language different than the first language. The method also includes generating, using a text-to-speech (TTS) model, an output audio feature representation of the input text by processing the input text sequence and the speaker embedding. The output audio feature representation includes the voice characteristics of the target speaker specified by the speaker embedding.

Abstract: A head mounted wearable computing device includes a frame including arm portions rotatably coupled to a front frame portion, with electronic components housed in and/or on the front frame portion and the arm portions. Cables run through hinge mechanisms rotatably coupling the arm portions and the front frame portion for connection of electronic components in the arm portions and the front frame portion. The hinge mechanism includes a dual pivot structure that guides the bending of cables through the hinge mechanism through approximately 90 degrees of bending through two pivot points. The dual pivot structure reduces localized strain on the cable through the range of motion of the hinge mechanism. The dual pivot structure allows for a rotation through the range of motion with little to no extra cable length to accommodate the bending of the cable.

Abstract: A user equipment (UE) employing different radio access technologies (RATs) concurrently provides the UE the opportunity to connect with different RAT-based base stations and concurrently transmit data thereto. A power-sharing control mechanism provides for sharing and allocating transmit power to multiple active RATs at the UE based on a priority designation of the data type associated with transmissions scheduled for each of the multiple active RATs. The power-sharing control mechanism provides efficient transmit power sharing between multiple transmit active RATs such that allocation of power to one RAT does not adversely affect the performance or coverage of the remaining RATs.

Abstract: Methods, systems, and apparatus, for totally corrective boosting with cardinality penalization are described. One of the methods includes obtaining initialization data identifying training examples, a dictionary of weak classifiers, and an active weak classifier matrix. Iterations of a totally corrective boosting with cardinality penalization process are performed, wherein each iteration performs operations comprising selecting a weak classifier from the dictionary of weak classifiers that most violates a constraint of a dual of the primal problem. The selected weak classifier is included in the active weak classifier matrix. The primal problem is optimized, and a discrete weight vector is determined. Weak classifiers are identified from the active weak classifier matrix with respective discrete weights greater than a threshold. The regularized risk is optimized, and a continuous weight vector is determined.

Abstract: Methods, apparatus, and computer-readable media for determining and utilizing corrections to robot actions. Some implementations are directed to updating a local features model of a robot in response to determining a human correction of an action performed by the robot. The local features model is used to determine, based on an embedding generated over a corresponding neural network model, one or more features that are most similar to the generated embedding. Updating the local features model in response to a human correction can include updating a feature embedding, of the local features model, that corresponds to the human correction. Adjustment(s) to the features model can immediately improve robot performance without necessitating retraining of the corresponding neural network model.

Abstract: This disclosure relates to digital video analysis. In one aspect, a method includes providing a user interface that enables a user of the computing system to select one or more seed video groups and one or more keywords, wherein each seed video group comprises one or more videos. The user interface is updated to provide candidate video groups selected based on the one or more seed video groups and the one or more keywords and, for each candidate video group, a first user interface control that enables the user to refine the set of candidate video groups to include video groups classified as being similar to the candidate video group. Data indicating user interaction with a given first user interface control for a first candidate video group is received. The user interface is updated to provide an updated set of candidate video groups.

Abstract: Systems and methods are provided for receiving at least one image and a reference image, and performing a plurality of downscaling operations having separable convolutions on the received at least one image. A plurality of residual blocks may be formed, with each residual block containing two separable convolutions of the kernel and two instance normalizations. A plurality of upscaling operations may be performed on the plurality of residual blocks, and a stylized image may be displayed based on at least the performed plurality of upscaling operations and the reference image.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for securely publishing and accessing a media's premium content. Methods include a media that comprises premium content encrypted with a first cryptographic key. The media also comprises data structures that include this key and the access entitlements. These data structures are encrypted using the key for each entity that can authorize user access to the media. Methods further include verifying user access to the media and then providing the decrypted premium content. Methods include decrypting the data structure using the authorizing entity's key and providing the first cryptographic key in the decrypted data structure to the client device, after verifying that the user has the appropriate access entitlements for the media. Methods include decrypting the premium content using the first cryptographic key and then displaying the media, including the decrypted premium content, on the client device.

Abstract: Text independent speaker recognition models can be utilized by an automated assistant to verify a particular user spoke a spoken utterance and/or to identify the user who spoke a spoken utterance. Implementations can include automatically updating a speaker embedding for a particular user based on previous utterances by the particular user. Additionally or alternatively, implementations can include verifying a particular user spoke a spoken utterance using output generated by both a text independent speaker recognition model as well as a text dependent speaker recognition model. Furthermore, implementations can additionally or alternatively include prefetching content for several users associated with a spoken utterance prior to determining which user spoke the spoken utterance.

Abstract: A method includes receiving training data that includes unspoken text utterances and un-transcribed non-synthetic speech utterances. Each unspoken text utterance is not paired with any corresponding spoken utterance of non-synthetic speech. Each un-transcribed non-synthetic speech utterance is not paired with a corresponding transcription. The method also includes generating a corresponding synthetic speech representation for each unspoken textual utterance of the received training data using a text-to-speech model. The method also includes pre-training an audio encoder on the synthetic speech representations generated for the unspoken textual utterances and the un-transcribed non-synthetic speech utterances to teach the audio encoder to jointly learn shared speech and text representations.

Abstract: The present disclosure provides systems and methods for content quasi-personalization or anonymized content retrieval via aggregated browsing history of a large plurality of devices, such as millions or billions of devices. A sparse matrix may be constructed from the aggregated browsing history, and dimensionally reduced, reducing entropy and providing anonymity for individual devices. Relevant content may be selected via quasi-personalized clusters representing similar browsing histories, without exposing individual device details to content providers.