Abstract: The technology relates to enhancing speech emotion recognition models with methods that enable the use of unlabeled data by inferring weak emotion labels. This is done by pre-trained large language models through weakly-supervised learning. For inferring weak labels constrained to a taxonomy, a textual entailment approach selects an emotion label with the highest entailment score for a speech transcript extracted via automatic speech recognition. The system may employ a method that generates, by one or more processors, a text transcript for a snippet of input speech, and then applies the text transcript to a pre-trained language model. The system can generate, using the pre-trained language model according to an engineered prompt and a predetermined taxonomy, a textual entailment from the text transcript. Based on this, the system may generate, by the one or more processors using the textual entailment, a predicted emotion corresponding to the input speech.

Abstract: A method includes receiving audio data corresponding to a query spoken and processing the audio data to generate multiple candidate hypotheses each represented by a respective sequence of hypothesized terms. For each candidate hypothesis, the method includes determining whether the sequence of hypothesized terms includes a source phrase from a list of phrase correction pairs. Each phrase correction pair includes a corresponding source phrase that was misrecognized and a corresponding target phrase replacing the source phrase. When the respective sequence of hypothesized terms includes the source phrase, the method includes generating a corresponding additional candidate hypothesis that replaces the source phrase.

Abstract: A method for securing messages includes obtaining, at a message server, a message for a user of a message service hosted by the message server. The message includes a header and the header includes a digital signature signed by an author of the message and a list of one or more recipients of the message. The method includes determining whether the digital signature by the author is valid and determining, using the list of one or more recipients, whether the user is a declared recipient of the message. When the digital signature by the author is valid and the user is the declared recipient of the message, the method includes delivering the message to a user device of the user. When the digital signature by the author is valid and the user is not the declared recipient of the message, the method includes alerting the user.

Abstract: Generating audio tracks is provided. The system selects a digital component object having a visual output format. The system determines to convert the digital component object into an audio output format. The system generates text for the digital component object. The system selects, based on context of the digital component object, a digital voice to render the text. The system constructs a baseline audio track of the digital component object with the text rendered by the digital voice. The system generates, based on the digital component object, non-spoken audio cues. The system combines the non-spoken audio cues with the baseline audio form of the digital component object to generate an audio track of the digital component object. The system provides the audio track of the digital component object to the computing device for output via a speaker of the computing device.

Abstract: Systems and methods for interactive playback of a video conference are provided. A request is received for a playback of a video conference between a plurality of participants of a plurality of client devices that each generated one of a plurality of source video streams, where each source video stream was presented during a live stream of the video conference according to a particular layout in a user interface (UI) on a first client device of the plurality of client devices.

Abstract: Systems and methods of verifying trigger keywords in acoustic-based digital assistant applications are provided. A system can receive, from an application developer computing device, a request to generate a voice-based software application. The request can include a uniform resource locator (URL) associated with a service provided by the voice-based software application and an identifier corresponding to the URL. The system can identify a plurality of links that include the URL as a target. The system can determine a subset of the plurality of links having a respective character string that includes a reference to the identifier for the URL. The data processing system can compare the subset of the plurality of links to a threshold to determine a match level, and can determine a verification status of the request based on the match level.

Abstract: A non-transitory computer-readable storage medium can include instructions stored thereon that, when executed by at least one processor, are configured to cause a computing device to determine, in response to a change in a refresh rate of a display, an encoded intensity of at least a portion of an image presented by the display, determine that the encoded intensity is within a predetermined range, and based on determining that the encoded intensity is within the predetermined range, adjust an intensity of a signal for the portion of the image.

Abstract: Implementations set forth herein relate to providing notifications regarding events that may interrupt content being rendered at an interface. The notifications can be preemptive and/or can indicate a predicted time and/or source for the events. The event can be, for example, a person attempting to contact a user who is viewing content at a display interface. The person can be associated with a food delivery that has been ordered by the user via a delivery application. An application, such an automated assistant application, can predict when the person is expected to arrive with the food delivery, and generate a notification ahead of the person arriving. In some implementations, the notification can be rendered at a scrubber user interface (UI) at a location corresponding to the time that the food delivery is expected to arrive, thereby putting the user on notice of when the streaming content may be interrupted.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating subqueries from a query. In one aspect, a method includes obtaining a query, generating a set of two subqueries from the query, where the set includes a first subquery and a second subquery, determining a quality score for the set of two subqueries, determining whether the quality score for the set of two subqueries satisfies a quality threshold, and in response to determining that the quality score for the set of two subqueries satisfies the quality threshold, providing a first response to the first subquery that is responsive to a first operation that receives the first subquery as input and providing a second response to the second subquery that is responsive to a second operation that receives the second subquery as input.

Abstract: Methods, systems and apparatus, including computer programs encoded on computer storage medium, for predicting a likelihood of a future computer memory failure. In one aspect training data inputs are obtained, where each training data input includes correctable memory error data that describes correctable errors that occurred in a computer memory and data indicating whether the correctable errors produced a failure of the computer memory. For each training data input, image representations of the correctable memory error data included in the training data input are generated. The image representations are processed using a machine learning model to output an estimated likelihood of a future failure of the computer memory. A difference between the estimated likelihood of the future failure of the computer memory and the data indicating whether the correctable errors produced a failure of the computer memory is computed. Values of model parameters are updated using the computed difference.

Abstract: Methods for creating a live copy of a data object from a production system for use by third party applications include receiving at least one request for a copy of production data from an application; creating a live backup copy; creating a flash copy of the live backup copy, and a flash copy bitmap; creating a modified version of the live backup copy by changing a subset of data in the live backup copy; recording the changed subset of data using the flash copy bitmap; mounting, the modified version of the live backup copy to the application; and transforming the modified version of the live backup copy back to the live backup copy when unmounting the modified version of the live backup copy of the production data from the application by applying changes associated with the flash copy bitmap to the live backup copy.

Abstract: Systems and methods for providing scene understanding can include obtaining a plurality of images, stitching images associated with the scene, detecting objects in the scene, and providing information associated with the objects in the scene. The systems and methods can include determining filter tags or query tags that can be selected to filter the plurality of objects, which can then be provided as information to the user to provide further insight on the scene. The information may be provided in an augmented-reality experience via text or other user-interface elements anchored to objects in the images.

Abstract: Systems and methods for generating min-increment counting bloom filters to determine count and frequency of device identifiers and attributes in a networking environment are disclosed. The system can maintain a set of data records including device identifiers and attributes associated with device in a network. The system can generate a vector comprising coordinates corresponding to counter registers. The system can identify hash functions to update a counting bloom filter. The system can hash the data records to extract index values pointing to a set of counter registers. The system can increment the positions in the min-increment counting bloom filter corresponding to the minimum values of the counter registers. The system can obtain an aggregated public key comprising a public key. The system can encrypt the counter registers using the aggregated shared key to generate an encrypted vector. The system can transmit the encrypted vector to a networked worker computing device.

Abstract: A method includes receiving a spoken utterance that includes a plurality of words, and generating, using a neural network-based utterance classifier comprising a stack of multiple Long-Short Term Memory (LSTM) layers, a respective textual representation for each word of the of the plurality of words of the spoken utterance. The neural network-based utterance classifier trained on negative training examples of spoken utterances not directed toward an automated assistant server. The method further including determining, using the respective textual representation generated for each word of the plurality of words of the spoken utterance, that the spoken utterance is one of directed toward the automated assistant server or not directed toward the automated assistant server, and when the spoken utterance is directed toward the automated assistant server, generating instructions that cause the automated assistant server to generate a response to the spoken utterance.

Abstract: A method for detecting network anomalies includes receiving a control message from a cellular network and extracting one or more features from the control message. The method also includes predicting a potential label for the control message using a predictive model configured to receive the one or more extracted features from the control message as feature inputs. Here, the predictive model is trained on a set of training control messages where each training control message includes one or more corresponding features and an actual label. The method further includes determining that a probability of the potential label satisfies a confidence threshold. The method also includes analyzing the control message to determine whether the control message corresponds to a respective network performance issue. When the control message impacts network performance, the method includes communicating the network performance issue to a network entity responsible for the network performance issue.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing highlight recovery. One of the methods includes receiving raw image data of an image, the raw image data comprising, for each of a plurality of highlight regions in the image, original channel values for one or more channels of the highlight region; processing the raw image data to generate updated image data, wherein the updated image data comprises updated channel values for each highlight region in the image, and wherein, for each of one or more highlight regions of the image, one or more updated channel values of the highlight region exceed a maximum channel value according to a predetermined data precision; and performing a hue correction process on the updated channel values of the one or more highlight regions of the image to generate final channel values that satisfy the predetermined data precision.

Abstract: This document describes a secure element that leverages the resources of a computer system to perform specialized functions using sensitive information. The secure element securely stores sensitive information on flash memory of the computer system. In response to a request requiring use of sensitive information, the secure element loads a security application and sensitive information from the computer system. By leveraging external resources, the secure element may flexibly accommodate increasing resource requirements of the computer system and be used in a wide range of computer systems.

Abstract: A trained model is retrained for video quality assessment and used to identify sets of adaptive compression parameters for transcoding user generated video content. Using transfer learning, the model, which is initially trained for image object detection, is retrained for technical content assessment and then again retrained for video quality assessment. The model is then deployed into a transcoding pipeline and used for transcoding an input video stream of user generated content. The transcoding pipeline may be structured in one of several ways. In one example, a secondary pathway for video content analysis using the model is introduced into the pipeline, which does not interfere with the ultimate output of the transcoding should there be a network or other issue. In another example, the model is introduced as a library within the existing pipeline, which would maintain a single pathway, but ultimately is not expected to introduce significant latency.

Abstract: Implementations set forth herein relate to generating a pre-call analysis for one or more users that are receiving and/or initializing a call with one or more other users, and/or prioritizing pre-call content according to whether security-related value was gleaned from provisioning certain pre-call content. One or more machine learning models can be employed for determining the pre-call content to be cached and/or presented prior to a user accepting a call from another user. Feedback provided before, during, and/or after the call can be used as a basis from which to prioritize certain content and/or sources of content when generating pre-call content for a subsequent call. Other information, such as contextual data (e.g., calendar entries, available peripheral devices, location, etc.) corresponding to the previous call and/or the subsequent call, can also be used as a basis from which to provide a pre-call analysis.