Abstract: The technology employs a patch-based multi-scale Transformer (300) that is usable with various imaging applications. This avoids constraints on image fixed input size and predicts the quality effectively on a native resolution image. A native resolution image (304) is transformed into a multi-scale representation (302), enabling the Transformer's self-attention mechanism to capture information on both fine-grained detailed patches and coarse-grained global patches. Spatial embedding (316) is employed to map patch positions to a fixed grid, in which patch locations at each scale are hashed to the same grid. A separate scale embedding (318) is employed to distinguish patches coming from different scales in the multiscale representation. Self-attention (508) is performed to create a final image representation. In some instances, prior to performing self-attention, the system may prepend a learnable classification token (322) to the set of input tokens.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: This document describes techniques and apparatuses for user-equipment coordination set (UECS) beam sweeping. In aspects, a user equipment (UE) receives an indication to coordinate beam sweeping with a UECS. The UE directs each UE in the UECS to perform a beam-training procedure by receiving a set of downlink beam transmissions, and forwards beam report information to a base station. In implementations, the UE receives an indication of one or more beam identities and one or more assigned time slots, and directs at least two UEs in the UECS to use specific beams indicated by the beam identities at specific time slots indicated by the assigned time slots, such as by transmitting a respective beam identity and a respective time slot to each UE of the at least two UEs.

Abstract: Systems and methods of dynamically varying the intensity of providing content items in a remarketing campaign based on tracking client device interactions are provided. The system can assign an account identifier to a first segment for a pre-conversion model, responsive to receiving a first interaction associated with a content provider from a client device. The system can assign the account identifier to a second segment for the pre-conversion model, responsive to receiving a second interaction. The system can assign the account identifier to a third segment, responsive to receiving a third interaction. The third interaction can include a conversion event. The system can generate a post-conversion model based on the third segment and the pre-conversion model. The system can determine an intent index for the account identifier based on the post-conversion model. The system can store the account identifier into an interest cluster based on the intent index.

Abstract: A first device provides both power and data to a second device over a power line connection between the two devices. The first device includes a power line extending from a power supply, a ground line extending from a ground, a first impedance in the power line, and a second impedance in the ground line. A modulator comprised of a transistor and modulator impedance is between the first impedance and the second impedance, and a tank capacitor is between the power line and the ground line, outside the first impedance and second impedance. A comparator is coupled between the first and second impedance. A switch may be included to short out the first and second impedance, thereby enabling transmission of only power for period of time, and return to a mode of transmitting both data and power. The first device may also receive data from the second device over the power line connection.

Abstract: A method includes establishing digital communication between a first user device and a second user device using a first codec. The method also includes selecting, based on an input signal representing an estimated unfiltered available bandwidth for the digital communication satisfying a first filter selection threshold, a first filter of two or more filters, and filtering the input signal using the first filter. The method further includes determining that the filtered input signal satisfies a first channel bandwidth threshold and, in response to determining that the filtered input signal satisfies the channel bandwidth threshold, selecting a second codec different from the first codec for further digital communication between the first user device and the second user device.

Abstract: An example operation may include a system comprising one or more of receiving a heartbeat failure notification in a VNFCI when the VNFCI is in standby state, sending to a Virtual Network Function Manager (VNFM), by an operational state machine, a next state request message, determining if a peer VNFCI is online when an administrative state of the peer VNFCI is online, determining an operational state of the peer VNFCI when the peer VNFCI is online, sending a first next state response message with a standby state to the VNFCI when the peer VNFCI operational state is active, sending a second next state response with an active state to the VNFCI when the peer VNFCI operational state is not active, examining, in the VNFCI, a next state attribute in a received next state response message, staying in a standby state when the next state attribute is standby, and transitioning to active state when the next state attribute is active.

Abstract: Various arrangements of electronic devices, such as assistant devices, are detailed herein. Such a device can include a housing and a microphone. A microphone aperture can be defined by the housing that directs sound from outside the housing to the microphone. A cover, which can be fabric, can be attached with the outside of the housing. An adhesive ring on the housing around the microphone aperture can be used to attach the cover to the housing.

Abstract: A method includes receiving an input audio signal corresponding to utterances spoken by multiple speakers. The method also includes encoding the input audio signal into a sequence of T temporal embeddings. During each of a plurality of iterations each corresponding to a respective speaker of the multiple speakers, the method includes selecting a respective speaker embedding for the respective speaker by determining a probability that the corresponding temporal embedding includes a presence of voice activity by a single new speaker for which a speaker embedding was not previously selected during a previous iteration and selecting the respective speaker embedding for the respective speaker as the temporal embedding. The method also includes, at each time step, predicting a respective voice activity indicator for each respective speaker of the multiple speakers based on the respective speaker embeddings selected during the plurality of iterations and the temporal embedding.

Abstract: A smart-home device may include a main power rail that provides power to components of the smart-home device; an integrator coupled to the main power rail that stores energy on an energy-storage device, where the energy stored on the energy-storage device is representative of an amount an amount of power provided to the smart-home device through the main power rail during an integration cycle of the integrator; and a counter that stores a number of integration cycles performed by the integrator during a time interval, where a total amount of power provided to the smart-home device through the main power rail during the time interval is represented by: (1) the number of integration cycles performed by the integrator during the time interval; and (2) the energy stored on the energy-storage device.

Abstract: A method includes receiving an automated speech recognition (ASR) request from a user device that includes a speech input captured by the user device and content metadata associated with the speech input. The content metadata is generated by the user device. The method also includes determining a priority score for the ASR request based on the content metadata associated with the speech input and caching the ASR request in a pre-processing backlog of pending ASR requests each having a corresponding priority score. The pending ASR requests in the pre-processing backlog are ranked in order of the priority scores. The method also includes providing, from the pre-processing backlog, one or more of the pending ASR requests to a backend-side ASR module, wherein pending ASR requests associated with higher priority scores are processed before pending ASR requests associated with lower priority scores.

Abstract: A method for training a streaming automatic speech recognition student model includes receiving a plurality of unlabeled student training utterances. The method also includes, for each unlabeled student training utterance, generating a transcription corresponding to the respective unlabeled student training utterance using a plurality of non-streaming automated speech recognition (ASR) teacher models. The method further includes distilling a streaming ASR student model from the plurality of non-streaming ASR teacher models by training the streaming ASR student model using the plurality of unlabeled student training utterances paired with the corresponding transcriptions generated by the plurality of non-streaming ASR teacher models.

Abstract: A method includes receiving training data that includes a set of unspoken textual utterances. For each respective unspoken textual utterance, the method includes, tokenizing the respective textual utterance into a sequence of sub-word units, generating a first higher order textual feature representation for a corresponding sub-word unit tokenized from the respective unspoken textual utterance, receiving the first higher order textual feature representation generated by a text encoder, and generating a first probability distribution over possible text units. The method also includes training an encoder based on the first probability distribution over possible text units generated by a first-pass decoder for each respective unspoken textual utterance in the set of unspoken textual utterances.

Abstract: A single E2E multitask model includes a speech recognition model and an endpointer model. The speech recognition model includes an audio encoder configured to encode a sequence of audio frames into corresponding higher-order feature representations, and a decoder configured to generate probability distributions over possible speech recognition hypotheses for the sequence of audio frames based on the higher-order feature representations. The endpointer model is configured to operate between a VAD mode and an EOQ detection mode. During the VAD mode, the endpointer model receives input audio frames, and determines, for each input audio frame, whether the input audio frame includes speech. During the EOQ detection mode, the endpointer model receives latent representations for the sequence of audio frames output from the audio encoder, and determines, for each of the latent representation, whether the latent representation includes final silence.

Abstract: Systems and techniques are provided for determination of user presence and absence using WiFi connections. Reports may be received from WiFi access points in an environment. The reports may include an identifier of a WiFi device, an indication of a connection to or disconnection from a WiFi access point, a time of the connection or disconnection, and an identifier of the WiFi access point. A connection sequence for the WiFi device may be generated from the reports. Whether the WiFi device is present in or absent from the environment as of a specified time may be determined based on the connection sequence. An indication of presence for a user associated with the WiFi device may generated if the WiFi device is present in the environment. An indication of absence for the user associated with the WiFi device may be generated if the WiFi device is absent from the environment.

Abstract: A method includes receiving training data that includes unspoken textual utterances in a target language. Each unspoken textual utterance not paired with any corresponding spoken utterance of non-synthetic speech. The method also includes generating a corresponding alignment output for each unspoken textual utterance using an alignment model trained on transcribed speech utterance in one or more training languages each different than the target language. The method also includes generating a corresponding encoded textual representation for each alignment output using a text encoder and training a speech recognition model on the encoded textual representations generated for the alignment outputs. Training the speech recognition model teaches the speech recognition model to learn how to recognize speech in the target language.

Abstract: A device may identify a plurality of sources for outputs that the device is configured to provide. The plurality of sources may include at least one of a particular application in the device, an operating system of the device, a particular area within a display of the device, or a particular graphical user interface object. The device may also assign a set of distinct voices to respective sources of the plurality of sources. The device may also receive a request for speech output. The device may also select a particular source that is associated with the requested speech output. The device may also generate speech having particular voice characteristics of a particular voice assigned to the particular source.

Abstract: A speaker verification method includes receiving audio data corresponding to an utterance, processing the audio data to generate a reference attentive d-vector representing voice characteristics of the utterance, the evaluation ad-vector includes ne style classes each including a respective value vector concatenated with a corresponding routing vector. The method also includes generating using a self-attention mechanism, at least one multi-condition attention score that indicates a likelihood that the evaluation ad-vector matches a respective reference ad-vector associated with a respective user. The method also includes identifying the speaker of the utterance as the respective user associated with the respective reference ad-vector based on the multi-condition attention score.

Abstract: A method includes processing, using a speech recognizer, a first portion of audio data to generate a first lattice, and generating a first partial transcription for an utterance based on the first lattice. The method includes processing, using the recognizer, a second portion of the data to generate, based on the first lattice, a second lattice representing a plurality of partial speech recognition hypotheses for the utterance and a plurality of corresponding speech recognition scores. For each particular partial speech recognition hypothesis, the method includes generating a corresponding re-ranked score based on the corresponding speech recognition score and whether the particular partial speech recognition hypothesis shares a prefix with the first partial transcription.

Abstract: Techniques and apparatuses are described that implement a smart-device-based radar system capable of performing angular position estimation. A machine-learned module analyzes complex range data generated to estimate angular positions of objects. The machine-learned module is implemented using a multi-stage architecture. In a local stage, the machine-learned module splits the complex range data into different range intervals and separately processes subsets of the complex range data using individual branch modules. In a global stage, the machine-learned module merges the feature data generated from the individual branch modules using a symmetric function and generates angular position data. By using machine-learning techniques and processing the complex range data directly, the radar system can achieve higher angular resolutions compared to other radar systems that utilize other techniques, such as analog or digital beamforming.