Abstract: A method for turn detection in a speech-to-speech model includes receiving, as input to the speech-to-speech (S2S) model, a sequence of acoustic frames corresponding to an utterance. The method further includes, at each of a plurality of output steps, generating, by an audio encoder of the S2S model, a higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames, and determining, by a turn detector of the S2S model, based on the higher order feature representation generated by the audio encoder at the corresponding output step, whether the utterance is at a breakpoint at the corresponding output step. When the turn detector determines that the utterance is at the breakpoint, the method includes synthesizing a sequence of output audio frames output by a speech decoder of the S2S model into a time-domain audio waveform of synthesized speech representing the utterance spoken by the user.

Abstract: Various arrangements for measuring blood pressure using sensor fusion are presented herein. Radio frequency (RF) signals are emitted and RF reflection signals are received by a radar sensor of a stationary device. The RF reflection signals are analyzed at a first distance range to identify a first pulse pressure waveform (PPW) at an aortic valve of a user. Vital sign data measured at an extremity of the user by a mobile device are received by the stationary device. The vital sign data is analyzed to identify a second PPW at the extremity of the user. Using the first PPW and the second PPW, a pulse transit time from the aortic valve to the extremity is determined. Using the PTT, a blood pressure (BP) of the user is determined and an indication of the BP is output.

Abstract: A method includes receiving a text input including a sequence of words represented as an input encoder embedding. The input encoder embedding includes a plurality of tokens, with the plurality of tokens including a first set of grapheme tokens representing the text input as respective graphemes and a second set of phoneme tokens representing the text input as respective phonemes. The method also includes, for each respective phoneme token of the second set of phoneme tokens: identifying a respective word of the sequence of words corresponding to the respective phoneme token and determining a respective grapheme token representing the respective word of the sequence of words corresponding to the respective phoneme token. The method also includes generating an output encoder embedding based on a relationship between each respective phoneme token and the corresponding grapheme token determined to represent a same respective word as the respective phoneme token.

Abstract: A method includes receiving one or more past trajectories navigated by a robotic device in an environment, wherein the one or more past trajectories are associated with initial environmental sensor data and one or more obstacle detection heuristics. The method also includes determining, based at least on subsequent environmental sensor data, one or more updated obstacle detection heuristics. The method further includes determining, based on the one or more updated obstacle detection heuristics and the initial environmental sensor data, one or more predicted drivable areas in the environment. The method additionally includes, based on the one or more predicted drivable areas including the one or more past trajectories, using the one or more updated obstacle detection heuristics to determine future navigation of the robotic device.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output temporal sequence of data elements conditioned on an input. In one aspect, a method comprises: obtaining the input, wherein the input comprises a noise input comprising a respective latent representation for each of a plurality of segments of the temporal sequence; updating, for each segment, the latent representation for the segment using a latent denoising neural network, the updating comprising, for each segment other than the first segment: obtaining a memory vector representing one or more hidden states generated by the latent denoising neural network when updating the latent representations for one or more preceding segments; updating the latent representation for the segment at each of a plurality of iterations; and generating the output temporal sequence of data elements by processing the latent representations for the plurality of segments.

Abstract: A system and method are disclosed. Audio input comprising the mixed audio signals is received by one or more client devices. The audio input is converted into a plurality of discrete tokens. A plurality of sound sources, each corresponding to a subset of discrete tokens of a plurality of subsets of discrete tokens, is determined using a trained machine learning model.

Abstract: A request of a user for a content item may be received, the user being associated with at least one of a set of preferences or a context. A group of similar users may be identified based the set of preferences or the context. A number of candidate content items corresponding to the content item may be identified. A difference in a measure of interaction of a first portion of the group after being provided the candidate content item and a second portion after not being provided the candidate content item may be determined for each candidate content item of the number of candidate content items. A subset may be determined based on the difference associated with each candidate content items. A candidate content item, identified based on at least one of the difference, an activity rate, or the context, may be provided to the user.

Abstract: Systems and methods are described that include operations such as detecting a plurality of computing devices configured as a distributed ambient computing system, receiving a request to execute a computing task, obtaining, from the distributed ambient computing system, data representing a device context for at least two of the plurality of devices, generating a combined context corresponding to the distributed ambient computing system, the combined context representing a combination of the device context for the at least two devices, generating and providing at least one decision request based on the computing task and the combined context, receiving a response to the at least one decision request, and triggering execution of the computing task based on the response and the combined context.

Abstract: This document discloses aspects of enforcing compliance of data use policies. In aspects, a network function receives, from an application, a message that includes data to upload from a device. In response, the network function executes a compliance function associated with a type of the data in the message. The compliance function can assert a data use policy based on a state (e.g., populated) of a field of the data and determine whether a user of the device has accepted the data use policy asserted for the data in the message. The compliance function can allow the upload of the data from the device based on compliance with the data use policy or blocks the upload of the data based on a lack of compliance with the data use policy. By so doing, the compliance function can ensure that data uploaded from the device complies with user-accepted data use policies.

Abstract: A method includes identifying, by a processing device, a third-party service provider of a plurality of third-party service providers that is authorized by a user to access data associated with the user. A data privacy score is generated based on one or more privacy risk factors. The data privacy score is associated with the third-party service provider. The data privacy score is indicative of a level of protection and privacy the third-party service provider maintains with respect to the data. A user interface (UI) displaying at least the data privacy score associated with the third-party service provider is provided for presentation on a client device associated with the user.

Abstract: Systems and methods for augmenting data can leverage one or more machine-learned models and contextual attention data to provide more realistic and efficient data augmentation. For example, systems and methods for inpainting can leverage a machine-learned model to generate predicted contextual attention data and blend the predicted contextual attention data with obtained contextual attention data to determine replacement data for augmenting an image to replace one or more occlusions. The obtained contextual attention data can include user-guided contextual attention.

Abstract: Some implementations relate to an automated assistant that can provide assistant suggestions based on content that the user has requested for the automated assistant to render at a computing device. In some instances, an assistant suggestion can correspond to a portion of the content that is not being rendered at the computing device but that is part of the content that the user is accessing. In other instances, an assistant suggestion can be based on data that can indicate how one or more users may have interacted with their automated assistant when viewing the content. Therefore, an assistant suggestion can correspond to an assistant operation that may have been initialized by another user who had previously viewed the content. In some implementations, an arrangement of the assistant suggestions can be modified according to which part of the content a user is viewing and/or user distance.

Abstract: The disclosed technology may comprise a mechanism that provides address range based memory hints, based on software memory access patterns. The mechanism includes an interface that can be in the form of a new instruction (“mem_range_hint”) that includes as operands (i) the start of the virtual address (“start_vaddr”), (ii) memory region size (“size”), and (iii) a hint (“hint”) associated with the memory access pattern.

Abstract: Systems and methods for providing a uni-card interface can provide for an interactive dynamic interface for providing information to a user. The uni-card interface may include a multi-state interface that can provide increasingly more detailed information upon each interaction. For example, the uni-card interface can provide a first state with condensed information, a second state with expanded information, and a third state with immersive information.

Abstract: A computer-implemented method for determining a score indicative of mindfulness of a user is provided. The method includes obtaining heart rate variability data of the user during a guided breathing exercise in which the user inhales and exhales to mimic a respiration rate associated with mindfulness. The method includes filtering the heart rate variability data to generate filtered heart rate variability. The method includes determining a first standard deviation of interbeat intervals indicative of respiratory sinus arrythmia and included in a first segment of the filtered heart rate variability data that spans a discrete interval of time. The method includes determining a second standard deviation of all interbeat intervals included in the first segment of the filtered heart rate variability data. The method includes determining a score indicative of mindfulness of the user based on the first standard deviation and the second standard deviation.

Abstract: A method includes identifying, by a processing device, a data structure including cells to store data, wherein the cells are arranged within rows and columns of the initial data structure. A scrambled data structure is created in which at least a subset of cells of the initial data structure is rearranged such that each cell of the rearranged subset of cells is located in at least one of a different row or a different column of the initial data structure. Responsive to receiving a user request to access the data of the initial data structure, the processing device determines whether to provide the user with access to the initial data structure or the scrambled data structure.

Abstract: A method includes receiving, by a processing device, a request to provide an entity with access to data associated with a user. A reference to location of the data within a data store controlled by the user and an access key associated with the data store is generated. Upon receiving an indication of user approval of the request, a response package is sent to the entity. The response package includes the reference to the location of the data, an obscured version of the data, and the access key. The data is obtained from the data store using the reference to the data and the access key provided by the entity. The data is provided to the entity.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for enabling artificial intelligence to generate new images based on contextual data and to generate digital components based on the images. In one aspect, a method includes receiving one or more queries from a client device of a user. A digital component is selected based on the one or more queries. A customized digital component is generated by obtaining an image of an object corresponding to the selected digital component and generating, using a language model, an image editing prompt for editing the image based on digital component data related to the digital component and query data including the one or more queries and contextual data. The image and the image editing prompt are provided to an image editing model. An edited image is received and used to generate the customized digital component.

Abstract: This document describes systems and techniques for indexing binaries of a software package and detecting potentially harmful code within the software package using API-call n-grams. A computing device generates API-call graphs from binaries. The computing device computes n-grams from the API-call graphs and adds them to an inverted index, which maps the n-grams to a respective identifier. The computing device identifies, using a signature that represents the behavior of the potentially harmful code, a set of candidate API-call graphs. The computing device can then compare, using a matching algorithm, the set of candidate API-call graphs to a non-deterministic finite automaton representation of the potentially harmful code. In this way, the described systems and techniques can use API-call n-grams to efficiently identify whether the software package includes a file that matches the behavior of potentially harmful code.

Abstract: A method includes receiving a prompt including a textual diarization request and corresponding audio data characterizing a conversation between multiple speakers. The method also includes generating a sequence of audio encoding chunks based on the corresponding data. For each respective audio encoding chunk, the method includes using a trained large language model (LLM) generating corresponding diarization results based on the respective audio encoding chunk and the textual diarization request and generating a new audio cohort for the respective audio encoding chunk based on the corresponding diarization results. The corresponding diarization results include a speech recognition result that has one or more predicted terms. Each respective predicted term is associated with a corresponding speaker token representing a predicted identity of a respective speaker that spoke the respective predicted term.

Abstract: A method includes receiving, at a computing device, a transcription of an audio recording, where the transcription is divided into a plurality of text sections. The method additionally includes associating one or more tags with each text section of the plurality of text sections, where each associated tag is extracted from text of the transcription. The method also includes selecting a subset of the plurality of text sections based at least on the one or more tags associated with each text section of the plurality of text sections. The method further includes providing, for display on the computing device, a representation of the transcription and a scroller interface, where the scroller interface allows for navigation to each text section in the subset based on user navigation to a tag associated with the text section.

Abstract: A method of generating high-performance images includes generating, by one or more processors, a first plurality of captions each corresponding to a different one of a first plurality of images. Generating the first plurality of captions includes inputting the first plurality of images into a first generative artificial intelligence (AI) model. The method also includes training or finetuning, by the one or more processors, a second generative AI model using the first plurality of images and the first plurality of captions, and generating, by the one or more processors, a second plurality of images. Generating the second plurality of images includes inputting a plurality of text prompts into the trained or finetuned second generative AI model.

Abstract: A method includes receiving a text embedding generation query from a user requesting generation of a text embedding for one or more data elements stored at a data warehouse. In response, the method includes selecting, using the text embedding generation query, a text embedding model from a plurality of different text embedding models. The method includes generating, using the selected text embedding model, the text embedding for the one or more data elements and storing the text embeddings at the data warehouse. The method includes receiving a machine learning model training query from the user device requesting training of a machine learning model using the text embeddings. In response to receiving the machine learning model training query, the method includes training the machine learning model using the text embeddings. The method includes providing, to the user device, a notification indicating that training of the machine learning model is complete.

Abstract: A method of optimizing a probability of precipitation forecast is provided. The method includes uniquely associating, based on a data segment definition, a spatiotemporal subregion with a probability of precipitation forecasting algorithm of two or more probability of precipitation forecasting algorithms. The method further includes selecting climatological data meeting the data segment definition and calibrating, using the selected climatological data meeting the data segment definition, the probability of precipitation forecasting algorithm associated with the spatiotemporal subregion.

Abstract: Moving large amounts of data from persistent storage to volatile memory involves significant data routing overhead that increases latency and consumes battery power. If an electronic device transitions to a deep sleep state to save power, restoring the system state data upon awakening is a slow process that negatively impacts device responsiveness. This document describes hardware and techniques that enable memory and storage control coordination. To accelerate data transfers, a local interconnect couples a storage controller to a memory controller. This arrangement allows the storage controller to bypass a coherent interconnect when sending data to the memory controller. By bypassing the coherent interconnect, the electronic device can avoid packetizing data into flits. Consequently, data transfer latency is reduced, enabling the electronic device to transition into aggressive power-saving states to extend battery life in a wider variety of scenarios.

Abstract: A method includes obtaining a transcript of multiple dialog acts during a multi-party communication session and extracting, from the transcript, a plurality of extractive action items each associated with a task expected to be completed within a timeframe to begin after the multi-party communication session ends. The method also includes generating, using an abstractive action item identification model configured to receive the extractive action items extracted from the transcript, one or more abstractive action items. Each abstractive action item is associated with a respective group of one or more of the extractive action items that are associated with a same respective task. For each of the one or more abstractive action items, the method also includes presenting, in one or more graphical user interfaces, information related to the corresponding abstractive action item.

Abstract: A method includes receiving a biased transcription for a voice command spoken by a user and captured by a user device, the biased transcription biased to include a biasing phrase from a set of biasing phrases specific to the user. The method also includes instructing an application executing on the user device to perform an action specified by the biased transcription for the voice command, and receiving one or more user behavior signals responsive to the application performing the action specified by the biased transcription. The method further includes generating, as output from a confidence model, a confidence score of the biased transcription based on the one or more user behavior signals input to the confidence model and, based on the confidence score output from the confidence model, training a speech recognizer on the biased transcription.

Abstract: The present disclosure describes techniques and apparatuses that are directed to integrated circuit design and fabrication. Designers may, for example, apply the techniques to identify failing Electrostatic Discharge (ESD) clamps at a floor planning stage, rather than waiting for signoff. These techniques, including a new flow process, may successfully identify violations missed by traditional flow processes, and may identify such violations earlier in the overall design and fabrication process.

Abstract: A method includes obtaining a user input interaction that selects a first graphical representation displayed on a graphical user interface (GUI). The method includes retrieving an application definition based on the user input interaction. The application definition includes a schema and an application template of the no-code application. The method includes retrieving contextual information associated with the no-code application. The method includes generating a usage prompt to summarize a usage history of the no-code application based on the application definition and the contextual information. The method includes generating a usage response based on the usage prompt. The usage response includes a summary of the usage history of the no-code application. The method includes transmitting, to the GUI, a second graphical representation of the summary of the usage history of the no-code application.

Abstract: The technology is generally directed to a gripping system for receiving hard drives of varying thicknesses and located in any position or orientation. The gripping system may include two arms to grip a hard drive and at least one belt positioned along a length of each arm to move the hard drive towards or away from the system. The arms may be pivotally attached so that hard drives of varying widths can be received between the arms. The belts may provide improved control of the hard drive once received between the arms because a force exerted by the belts may secure the hard drive in place to prevent the hard drive from drooping or moving out of place.

Abstract: Implementations described herein relate to a computer-implemented method that includes capturing image data using one or more cameras, wherein the image data includes a primary image and associated depth values. The method further includes encoding the image data in an image format. The encoded image data includes the primary image encoded in the image format and image metadata that includes a device element that includes a profile element indicative of an image type and a first camera element, wherein the first camera element includes an image element and a depth map based on the depth values. The method further includes, after the encoding, storing the image data in a file container based on the image format. The method further includes causing the primary image to be displayed.

Abstract: The present disclosure describes various aspects of implementing parallelizing secure monitor calls in an Arm architecture. In some aspects, an apparatus receives, from an application in a non-secure zone of the computing system, a Secure Monitor Call (SMC) requesting use of one or more trusted resources in a secure zone of the computing system. The apparatus enqueueing the SMC as an entry in a receive (RX) queue and evaluates SMC call information to schedule the SMC call for execution using the one or more trusted resources in the secure zone. Based on the execution of the SMC call, a scheduler receives a response to the scheduled SMC call. The apparatus enqueues the response to the SMC call in a transmit (TX) queue and sends a notification, to the application, indicating that the response to the SMC call is in the TX queue.

Abstract: Techniques for determining perceptual quality indicators of video content items are provided. In some embodiments, a system including one or more processors executes instructions to: receive a video content item comprising a plurality of frames; determine, using a first subnetwork of a deep neural network, a content quality indicator for each frame, wherein the content quality indicator corresponds to one or more semantic content indicators for each frame; determine, using a second subnetwork of the deep neural network, a video distortion indicator for each frame, wherein the video distortion indicator indicates a quality of the frame based on distortions contained within the frame; generate a quality level for each frame based on at least the content quality indicator and the video distortion indicator for the frame; and output an indication of quality for the video content item based on the quality level for each frame.

Abstract: A user device receives a request to generate a composite image. The media application a first image that includes one or more first subjects. The media application determines a previous pose of the user device associated with capture of the first image. The media application segments the one or more first subjects from the first image. The media application generates one or more overlays that correspond to the one or more first subjects based on segmenting the one or more first subjects. The media application displays the one or more overlays on a viewfinder of the user device to provide guidance for a user to capture a second image based on a comparison of a current pose of the user device to the previous pose of the user device. The media application generates the composite image.

Abstract: A display system may include a segment analyzer, a segment buffer, and a set of pixel drivers. The segment analyzer may perform an analysis of image data representing an image to be displayed. The analysis distinguishes a first portion of the image data that corresponds to an active segment of the image containing non-zero content from a second portion of the image data that corresponds to an empty segment of the image without non-zero content. The segment buffer stores the first portion of the image data in accordance with a segment map based on the analysis. The pixel drivers then use the segment map and the image data to cause a set of pixels to display the image by driving a first subset of pixels associated with the active segment without driving a second subset of pixels associated with the empty segment.

Abstract: The present disclosure provides systems, methods, and computer program products for providing efficient embedding table storage and lookup in machine-learning models.

Abstract: Systems and methods for real-time image deblur and stabilization can utilize sensor data for estimating motion blur without the high computational cost of image analysis techniques. The estimated motion blur can then be utilized to generate a motion blur kernel for image correction. The systems and methods can further refine the correction by processing the motion blur kernel with a polynomial filter to generate a sharpening kernel. The systems and methods can provide for real-time correction even with minimal to no stabilization masking.

Abstract: A media application detects a bystander in an initial image. The media application generates a bystander box that includes the bystander, wherein all pixels for the bystander are within the bystander box. The media application generates localizer boxes that encompass the bystander and one or more objects that are attached to the bystander. The media application aggregates the bystander box and one or more of the localizer boxes to form an aggregated box. The media application applies a segmenter to the initial image, based on the aggregated box, to segment the bystander and the one or more objects from the initial image to generate a bystander mask, wherein the bystander mask includes a subset of pixels within the aggregated box. The media application generates an inpainted image that replaces all pixels within the bystander mask with pixels that match a background in the initial image.

Abstract: A light-emitting diode (LED) structure includes an active region that has at least one aluminum-containing quantum well (QW) stack that emits light from the LED structure when activated. The LED structure exhibits a modified internal quantum efficiency value, which is higher than a LED structure that does not include aluminum within a QW stack. The LED structure also exhibits a modified peak wavelength, which is longer than an unmodified peak wavelength of the unmodified LED structure.

Abstract: An example technique for image analysis is provided. An example image analysis method includes obtaining an instructive sequence descriptive of an instructive query, an instructive response, and an instructive trace of intermediate states from the instructive query to the instructive response. The example image analysis method includes inputting, to a machine-learned model, the instructive sequence and an operative image processing query that comprises image data, wherein the machine-learned model is configured to process the operative query with attention over the instructive sequence. The example method can include generating, using the machine-learned model and responsive to the operative query, an operative image processing response that comprises an analysis of the image data.

Abstract: Some implementations receive audio data capturing a verb-based succinct query that includes an action but is void of any app entity and determine, based on processing of the audio data, the action and one or more instance candidates associated with the action. Some of those implementations further query an app entity database for usage information and/or capability information of one or more app entities that correspond to the one or more instance candidates and generate, based on selecting from the one or more app entities, a response or action responsive to the verb-based succinct query.

Abstract: To determine whether a web resource complies with a policy for displaying an informational notice, a system obtaining a Document Object Model (DOM) representation of the web resource, the DOM representation specifying a hierarchy of pages and including instructions for displaying content elements. The system traverses the DOM representation of the web resource to determine whether the web resource includes the informational notice, including calculating a likelihood that a content element corresponds to the informational notice based on one or more keywords associated with the informational notice. In response to determining that the content element corresponds to the informational notice, the system determines a set of visual parameters of the content element, and determines whether the web resource is configured to display the content element according to a visibility metric.

Abstract: A method for providing elastic columnar cache includes receiving cache configuration information indicating a maximum size and an incremental size for a cache associated with a user. The cache is configured to store a portion of a table in a row-major format. The method includes caching, in a column-major format, a subset of the plurality of columns of the table in the cache and receiving a plurality of data requests requesting access to the table and associated with a corresponding access pattern requiring access to one or more of the columns. While executing one or more workloads, the method includes, for each column of the table, determining an access frequency indicating a number of times the corresponding column is accessed over a predetermined time period and dynamically adjusting the subset of columns based on the access patterns, the maximum size, and the incremental size.

Abstract: A joint segmenting and ASR model includes an encoder to receive a sequence of acoustic frames and generate, at each of a plurality of output steps, a higher order feature representation for a corresponding acoustic frame. The model also includes a decoder to generate based on the higher order feature representation at each of the plurality of output steps a probability distribution over possible speech recognition hypotheses, and an indication of whether the corresponding output step corresponds to an end of segment (EOS).

Abstract: The technology is directed to artificial intelligence (AI) powered tools that can enhance existing digital video components and simplify and automate the creation of new digital video components. The technology includes a digital video component creation tool that leverages existing assets to generate digital video components, a voice-over tool that can add voice-overs, generated from text to video components, and a video component evaluation tool that can evaluate video components for conformity with attributes associated with metrics for video creatives.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for verifying current transformer polarity for one or more current transformers. In one aspect, a method comprises for each input power leg of a protected device, injecting a first test current into a first primary node H1 of a respective current transformer to induce a second test current in a first secondary node S1 of the respective current transformer, determining a positive test result based on the flow of the second current, and determining a polarity verification for each of the input power legs of the protected device when a positive test result is determined for each of the one or more input legs of the protected device.

Abstract: According to an aspect, a method includes executing a first application as a foreground process of an operating system of a computing device, and displaying, by the first application, content on a display of the computing device, where the content is associated with an action executable by a second application. The method includes transmitting, by the first application, an initialization request to an initialization application programming interface (API) of the second application in advance of a user action taken with respect to the content displayed on the display, where the initialization request is configured to cause the second application to be launched as a background process of the operating system of the computing device.

Abstract: Implementations relate to providing a user feedback mechanism that enables a user to provide feedback towards one or more specific portions of a response. The response can be generated based on processing of a user query using a generative model such as a large language model (LLM). The one or more specific portions can be a textual portion that includes textual content, and/or a media content portion that include media content such as one or more images, one or more videos, one or more audio pieces, etc. The feedback towards one or more of the specific portions of the response can be utilized in training or fine-tuning the generative model (or an additional generative model) via approaches such as supervised training or reinforced learning.