Abstract: A computer-implemented technique can include detecting, by a first computing device, a set of user communications at least one of transmitted to and received by from a second computing device via a first communication mode, identifying a second communication mode that is available for communication between the first and second computing devices, and obtaining an appropriateness score for the first and second communication modes based on a contextual feature of the set of user communications, wherein the contextual feature relates an appropriateness of a particular communication mode for the set of user communications, and wherein each appropriateness score is indicative of a level of the appropriateness of a particular communication mode for the set of user communications. The technique can also include selectively outputting a suggestion to switch from the first communication mode to the second communication mode.
Type:
Grant
Filed:
March 10, 2020
Date of Patent:
April 12, 2022
Assignee:
GOOGLE LLC
Inventors:
Matthew Sharifi, Jakob Nicolaus Foerster
Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for disambiguating join paths for natural language queries. One of the methods includes, obtaining a natural language query from a user; parsing the natural language query into structured operations to be performed on APIs of a knowledge base, including: responsive to detecting a parsing ambiguity in which the natural language query can be parsed in two or more ways: providing, through a user interface, to the user one or more information items identifying the parsing ambiguity; responsive to a user interaction with an information item: modifying the parsing in accordance with the user interaction to generate one or more structured operations; performing the one or more structured operations on the structured APIs of the knowledge base to determine one or more search results; and providing one or more search results to the user.
Abstract: At least one aspect of the present disclosure is directed to systems and methods of secure and privacy preserving device classification. A server can maintain a plurality of data records, each including an indication of a request and a known classification value. The server can train a context obfuscation model using each of the plurality of requests and known classification values. The server can train a classification model using resources and category information from a data structure in the memory of the client device. The server can transmit the context obfuscation model to a different plurality of client devices. The server can receive a request for classification including a classification vector and request metadata. The server can determine the classification of the device responsible for the request using the classification model. The server can transmit the device classification to the device responsible for the request.
Type:
Application
Filed:
April 3, 2020
Publication date:
April 7, 2022
Applicant:
GOOGLE LLC
Inventors:
Mahesh KERALAPURA MANJUNATHA, Chiu Wah SO
Abstract: Modulating packetized audio signals in a voice activated data packet based computer network environment is provided. A system can receive audio signals detected by a microphone of a device. The system can parse the audio signal to identify trigger keyword and request, and generate a first action data structure. The system can identify a content item object based on the trigger keyword, and generate an output signal comprising a first portion corresponding to the first action data structure and a second portion corresponding to the content item object. The system can apply a modulation to the first or second portion of the output signal, and transmit the modulated output signal to the device.
Type:
Grant
Filed:
July 22, 2019
Date of Patent:
April 5, 2022
Assignee:
GOOGLE, LLC
Inventors:
Gaurav Bhaya, Robert Stets, Bailiang Zhou
Abstract: A MapReduce-based training framework exploits both data parallelism and model parallelism to scale training of complex models. Particular model architectures facilitate and benefit from use of such training framework. As one example, a machine-learned model can include a shared feature extraction portion configured to receive and process a data input to produce an intermediate feature representation and a plurality of prediction heads that are configured to receive and process the intermediate feature representation to respectively produce a plurality of predictions. For example, the data input can be a video and the plurality of predictions can be a plurality of classifications for content of the video (e.g., relative to a plurality of classes).
Abstract: A processor decodes, from a compressed bitstream, an adaptive intra-prediction mode of a set of adaptive filter modes, the adaptive intra-prediction mode indicating a number of filter coefficients and relative locations with respect to a to-be-predicted pixel of a sub-set of neighboring pixels of the to-be-predicted pixel; determines filter coefficients for generating a prediction block of the block; and generates, by recursive extrapolations that use the filter coefficients and the relative locations, the prediction block. The set of adaptive filter modes includes a first adaptive mode and a second adaptive mode. The first adaptive mode and the second adaptive mode indicate a same number of coefficients. The first adaptive mode indicates a first set of first relative locations of a first sub-set of neighboring pixels that is different from a second set of second relative locations of a second sub-set of neighboring pixels indicated by the second adaptive mode.
Abstract: Techniques of source localization and acquisition involve a wideband joint acoustic source localization and acquisition approach in light of sparse optimization framework based on an orthogonal matching pursuit-based grid-shift procedure. Along these lines, a specific grid structure is constructed with the same number of grid points as compared to the on-grid case, but which is “shifted” across the acoustic scene. More specifically, it is expected that each source will be located close to a grid point in at least one of the set of shifted grids. The sparse solutions corresponding to the set of shifted grids are combined to obtain the source location estimates. The estimated source positions are used as side information to obtain the original source signals.
Type:
Grant
Filed:
October 10, 2018
Date of Patent:
April 5, 2022
Assignee:
GOOGLE LLC
Inventors:
Willem Bastiaan Kleijn, Jan Skoglund, Christos Tzagkarakis
Abstract: Techniques are disclosed for training and/or utilizing an automatic post-editing model in correcting translation error(s) introduced by a neural machine translation model. The automatic post-editing model can be trained using automatically generated training instances. A training instance is automatically generated by processing text in a first language using a neural machine translation model to generate text in a second language. The text in the second language is processed using a neural machine translation model to generate training text in the first language. A training instance can include the text in the first language as well as the training text in the first language.
Type:
Grant
Filed:
July 15, 2019
Date of Patent:
April 5, 2022
Assignee:
GOOGLE LLC
Inventors:
Markus Freitag, Isaac Caswell, Howard Scott Roy
Abstract: A method of measuring a banding artefact in an image includes generating a gradient profile from the image, where the gradient profile includes respective gradient magnitudes of pixels of the image; generating, using the gradient profile, a candidate banding pixel (CBP) map, where each location of the CBP map is such that a gradient magnitude of the gradient profile of a corresponding pixel of the image being greater than a first threshold and smaller than a second threshold; generating, using the CBP map, a banding edge map (BEM), where the BEM includes connected banding edges of the image; generating, using the BEM, a banding visibility map (BVM), where the BVM includes a respective banding metric for at least some pixels of the image; and generating a banding index of the image using the BVM.
Abstract: The present disclosure provides systems and methods for interrupting streaming content provided via a manifest inviolate protocol. An intelligent streaming server or edge cache may substitute different content than that which is requested, without the client's knowledge. In some implementations, the client may request a first segment of a file or stream, and the streaming server may instead deliver a segment of an entirely different file or stream. The replacement segment may have the same length as the requested segment, and may be renamed such that the client believes that the requested segment has been properly served. Accordingly, without changing the manifest or departing from the requirements of the manifest-inviolate protocol and without changing any functionality of the client, a system may provide content switching on a per-segment basis, rather than a per-manifest basis.