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: This document describes a vapor chamber within an electronic device. In aspects, an electronic device includes a middle frame that provides mechanical support for the electronic device, a middle plate affixed to the middle frame to define an inner layer of a chassis, and a vapor chamber disposed inside the middle plate. The vapor chamber includes a first region proximate to a heat source and a second region opposite the first region. A coolant is evaporated in a first mode at the first region by heat absorbed from the heat source and is condensed in a second mode in the second region. This vapor chamber permits cooling of elements within the electronic device at lower cost and/or smaller size than many conventional cooling systems.

Abstract: A method for a statistics collection framework includes receiving a schema defining a relational database for storing a plurality of statistics corresponding to a query, the relational database including a plurality of data tables relationally connected according to the schema, each data table of the plurality of data tables corresponding to a respective statistic. The method includes receiving a query corresponding to data at a data store. The method also includes executing the query. During execution of the query, the method includes collecting, from a query execution database, the plurality of statistics related to the query, each statistic of the plurality of statistics corresponding to a respective data table of the plurality of data tables of the relational database and, for each statistic of the plurality of statistics, storing the respective statistic at the respective data table according to the schema.

Abstract: A method includes receiving training data that includes a plurality of sets of training utterances each associated with a respective language. Each training utterance includes a corresponding reference speech representation paired with a corresponding input text sequence. For each training utterance, the method includes generating a corresponding encoded textual representation for the corresponding input text sequence, generating a corresponding speech encoding for the corresponding reference speech representation, generating a shared encoder output, and determining a text-to-speech (TTS) loss based on the corresponding encoded textual representation, the corresponding speech encoding, and the shared encoder output. The method also includes training a TTS model based on the TTS losses determined for the training utterances in each set of the training utterances to teach the TTS model to learn how to synthesize speech in each of the respective languages.

Abstract: A method includes receiving a sequence of acoustic frames characterizing a spoken utterance in a particular native language. The method also includes generating a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames by a causal encoder that includes an initial stack of multi-head attention layers. The method also includes generating a second higher order feature representation for a corresponding first higher order feature representation by a non-causal encoder that includes a final stack of multi-head attention layers. The method also includes receiving, as input at each corresponding language-dependent adapter (LDA) module, a language ID vector identifying the particular native language to activate corresponding language-dependent weights specific to the particular native language. The method also includes generating a first probability distribution over possible speech recognition hypotheses by a decoder.

Abstract: Various arrangements are presented for performing a hierarchical application launch of an application. A requests to register one or more smart home devices can be received in association with a user account. The smart home devices can be mapped to the user account based on receiving the requests. An application that is mapped to the user account may be launched; the application can analyze the one or more smart home devices registered to the user account and a user interface hierarchy. Based on analyzing the one or more smart home devices registered to the user account and the user interface hierarchy, an initial launch interface can be selected and output for presentation.

Abstract: A method includes receiving, by a processing device of a content sharing platform, a request for desired content from a client device, the content being stored in a content delivery network (CDN). The method further includes generating, based on data available to the content sharing platform, a partial trust metric associated with the client device, wherein the partial trust metric is to be used by a CDN server to make a decision regarding access to the desired content by the client device. The method further includes generating a response to the content request, wherein the response comprises one or more resource locators for accessing the desired content in the CDN, and the partial trust metric. The method further includes sending the response to the client device to enable the client device to request the desired content from the CDN server using the resource locator(s) and the partial trust metric.

Abstract: The present disclosure provides systems and method for adjusting the audio output of a wearable device based on an audio gain profile of a user. The wearable device may receive an audiogram indicating one or more frequency ranges associated with hearing loss. The wearable device may determine the audio gain profile based on the audiogram. The wearable device may use one or more adjustment modules to determine a gain to apply to the audio output based on the audio gain profile. The adjustment modules may include an active noise control module, a hearing assistance module, and a transparency control module. The wearable device may determine the amount of gain to apply using a least mean square algorithm and/or a machine learning model.

Abstract: Aspects of the disclosure are directed to monitoring, alerting, and/or root causing network problems based on current network behavior and network events at any instant in time using a network behavior model. The network behavior model can learn and be updated with network states and events to correlate network problems and determine root causes of the network problems for alerting and/or automatic correction.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium for retrieving information from a server. Methods can include a server receiving a set of client-encrypted queries. The server identifies a set of server-encrypted decryption keys and transmits the set to the client device. The server receives a set of client-server-encrypted decryption keys that includes the set of server-encrypted decryption keys encrypted by the client device. The server also receives a set of client-encrypted/client-derived decryption keys that were derived by the client device. The server generates matching a map that specifies matches between the set of client-server-encrypted decryption keys and the set of client-encrypted/client-derived decryption keys. The server filters the set of client-encrypted queries using the map to create a set of filtered client-encrypted queries and generates a set of query results.

Abstract: Aspects of the disclosure provide for a system for navigating a conversation graph using a language model trained to generate Application Programming Interface (API) calls in response to natural language input from a user computing device. A conversational agent implementing a state handler and a language model (LM) communicates with a user computing device through a user frontend. Rather than communicating directly with a user with output in natural language, the agent uses a (LM) trained as described herein to navigate a conversation graph. The state handler receives API calls generated by the LM and updates the state of a conversation with a user as indicated in the graph. After the update, the state handler can perform one or more predetermined actions associated with a node indicating the current state of the conversation.

Abstract: Implementations can reduce the time required to obtain responses from an automated assistant through proactive caching, locally at a client device, of proactive assistant cache entries—and through on-device utilization of the proactive assistant cache entries. Different proactive cache entries can be provided to different client devices, and various implementations relate to technique(s) utilized in determining which proactive cache entries to provide to which client devices. In some of those implementations, in determining which proactive cache entries to provide (proactively or in response to a request) to a given client device, a remote system selects, from a superset of candidate proactive cache entries, a subset of the cache entries for providing to the given client device.

Abstract: A system and method is provided that displays cursors for street level images, where the cursor changes appearance based on the objects in the image, such as the geographic distance between the objects and the camera position and the surface of the objects. For example, the cursor may appear to lie flat against the objects in the image change size based on the distance between the camera and object's surface.

Abstract: The technology is generally directed to a coordinated power throttling mechanism for a payload using power provided by a rack such that the rack power does not exceed a threshold amount for greater than a predetermined period of time. The coordinated power throttling mechanism includes the rack providing a power throttling signal to the payload and the payload executing the power throttling upon detection of the throttling signal. The payload may detect the throttling signal and, after a delay, execute the power throttling. The delay may ensure that all payloads within the rack have detected the power throttling signal.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for speech recognition using neural networks. A feature vector that models audio characteristics of a portion of an utterance is received. Data indicative of latent variables of multivariate factor analysis is received. The feature vector and the data indicative of the latent variables is provided as input to a neural network. A candidate transcription for the utterance is determined based on at least an output of the neural network.

Abstract: A system and method for protocol independent receive side scaling (RSS) includes storing a plurality of RSS hash M-tuple definitions, each definition corresponding to one of a set of possible protocol header combinations for routing an incoming packet, the set of possible protocol header combinations being modifiable to include later-developed protocols. Based on initial bytes of the incoming packet, a pattern of protocol headers is detected, and used to select one of the plurality of RSS hash M-tuple definitions. The selected RSS hash M-tuple definition is applied as a protocol-independent arbitrary set of bits to the headers of the incoming packet to form a RSS hash M-tuple vector, which is used to compute a RSS hash. Based on the RSS hash, a particular queue is selected from a set of destination queues identified for the packet, and the packet is delivered to the selected particular queue.

Abstract: Implementations described herein utilize an independent server for facilitating secure exchange of data between multiple disparate parties. The independent server receives client data, via an automated assistant application executing at least in part at a client device, that is to be transmitted to a given third-party application. The independent server processes the client data, using a first encoder-decoder model, to generate opaque client data, and transmits the opaque client data to the given third-party application and without transmitting any of the client data. Further, the independent server receives response data, via the given third-party application, that is generated based on the opaque client data and that is to be transmitted back to the client device. The independent server processes the response data, using a second encoder-decoder model, to generate opaque response data, and transmits the opaque response data to the client device and without transmitting any of the response data.

Abstract: A display device based on micro light emitting diodes (microLEDs) includes a plurality of chiplets. Each chiplet includes one or more raxels, each raxel including a plurality of microLEDs supported on a substrate. The chiplet also includes a micro integrated circuit (microIC) electronically connected with the one or more raxels. MicroIC includes a plurality of interconnects supported on a backplane such that, when connected with the raxel, microIC may be used to electrically drive each one of the MicroLED of the raxel. In an embodiment, a plurality of chiplets are disposed on a display substrate to for an auto-view horizontal parallax only 3D display.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for handing off a user conversation between computer-implemented agents. One of the methods includes receiving, by a computer-implemented agent specific to a user device, a digital representation of speech encoding an utterance, determining, by the computer-implemented agent, that the utterance specifies a requirement to establish a communication with another computer-implemented agent, and establishing, by the computer-implemented agent, a communication between the other computer-implemented agent and the user device.