Abstract: An audio-visual automated speech recognition model for transcribing speech from audio-visual data includes an encoder frontend and a decoder. The encoder includes an attention mechanism configured to receive an audio track of the audio-visual data and a video portion of the audio-visual data. The video portion of the audio-visual data includes a plurality of video face tracks each associated with a face of a respective person. For each video face track of the plurality of video face tracks, the attention mechanism is configured to determine a confidence score indicating a likelihood that the face of the respective person associated with the video face track includes a speaking face of the audio track. The decoder is configured to process the audio track and the video face track of the plurality of video face tracks associated with the highest confidence score to determine a speech recognition result of the audio track.

Abstract: A compact electronic device has a touch sensor and/or a microphone that are concealed within a housing at least partially wrapped by an acoustically porous cover. In some implementations, the touch sensor includes a sensing portion and a contact portion extending from the sensing portion. While the sensing portion is placed in proximity to an interior surface of the housing to detect a touch on the housing, the contact portion is bent to electrically couple the sensing portion to a circuit board via two distinct electrical paths. In some implementations, an exterior surface of the housing includes a sealing area surrounding an aperture on the housing, and the acoustically porous cover is affixed to the sealing area via an adhesive. The adhesive covers the sealing area and permeates a thickness of the acoustically porous cover, thereby enabling formation of a controlled sound path to access the microphone via the aperture.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for providing a machine learning model that is trained to perform a machine learning task. In one aspect, a method comprises receiving a request to train a machine learning model on a set of training examples; determining a set of one or more meta-data values characterizing the set of training examples; using a mapping function to map the set of meta-data values characterizing the set of training examples to data identifying a particular machine learning model architecture; selecting, using the particular machine learning model architecture, a final machine learning model architecture for performing the machine learning task; and training a machine learning model having the final machine learning model architecture on the set of training examples.

Abstract: A method of reducing junction resistance variation for junctions in quantum information processing devices fabricated using two-step deposition processes. In one aspect, a method includes providing a dielectric substrate, forming a first resist layer on the dielectric substrate, forming a second resist layer on the first resist layer, and forming a third resist layer on the second resist layer. The first resist layer includes a first opening extending through a thickness of the first resist layer, the second resist layer includes a second opening aligned over the first opening and extending through a thickness of the second resist layer, and the third resist layer includes a third opening aligned over the second opening and extending through a thickness of the third resist layer.

Abstract: The technology relates to a memory insertion apparatus for pushing memory modules into corresponding memory sockets on a circuit board. The memory insertion apparatus may include a frame and an actuation assembly coupled to the frame, and one or more cam assemblies each rotatably coupled to the frame and operatively coupled to the actuation assembly. Each cam assembly may include a central shaft extending in a longitudinal direction, and a plurality of cams each having a tip configured to engage one of the memory modules, the tip extending from the central shaft in a respective radial direction perpendicular to the longitudinal direction. A center of each cam may be offset from centers of adjacent ones of the cams by a pitch distance that is about equal to a pitch distance between centers of adjacent ones of the memory sockets.

Abstract: An apparatus for handling electronic components such as hard disk drives. In one aspect, the apparatus includes a main body defining an interior space with an open front; a drive system that propels and positions the apparatus along horizontal surface; a fan system mounted within the interior space and positioned to blow air down into the interior space; a first gripper apparatus that engages an equipment drawer of an electronics rack; and a second gripper apparatus that grips and removes a target electronic component from a target position located within the equipment drawer, wherein at least a back surface of the main body includes perforations so that sufficient air flow generated by the fan system flows through the perforations to maintain cooling of electronic components in the equipment drawer when the equipment drawer is in the extracted position.

Abstract: Methods, systems, and media for generating a group of media content items are provided. In some embodiments, the method comprises: receiving, from a server, a plurality of media content items; storing, on a user device, the plurality of media content items; determining, by the user device, whether to request a clustering of the plurality of media content items; in response to determining that a request for clustering the plurality of media content items is to be made, transmitting the request to the server, wherein the request includes indications of the plurality of media content items; receiving, from the server, a plurality of clusters corresponding to the plurality of media content items; and causing indications of the plurality of media content items to be presented in a user interface on the user device, wherein the indications of the plurality of media content items are organized based on the plurality of clusters.

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: The various implementations described herein include methods, devices, and systems for implementing high dynamic range and automatic exposure functions in a video system. In one aspect, a method is performed at a video camera device and includes, while operating in a non-high dynamic range (HDR) mode: capturing first video data of a scene with the image sensor; determining whether a minimum number of pixels of the first video data meets one or more first color intensity criteria; and in accordance with the determination that the minimum number of pixels of the first video data meets the one or more first color intensity criteria, switching operation from the non-HDR mode to an HDR mode.

Abstract: A method for searchable encryption with a public key includes receiving an operation request from a user device associated with a user requesting that encryption of data associated with the user. The data includes a corpus of documents stored on a remote storage device. The method also includes receiving a public key associated with the user. The public key includes an asymmetric cryptographic public key. The method also includes generating a random data key. The data key includes a symmetric cryptographic key. The method also includes encrypting, using the data key, a search index for the corpus of documents based on keywords within the corpus of documents. The method also includes encrypting, using the public key, the data key and sending the encrypted data key to a user device associated with the user.

Abstract: A method for generating an output image from an input image and an input text instruction that specifies a location and a modification of an edit applied to the input image using a neural network is described. The neural network includes an image encoder, an image decoder, and an instruction attention network. The method includes receiving the input image and the input text instruction; extracting, from the input image, an input image feature that represents features of the input image using the image encoder; generating a spatial feature and a modification feature from the input text instruction using the instruction attention network; generating an edited image feature from the input image feature, the spatial feature and the modification feature; and generating the output image from the edited image feature using the image decoder.

Abstract: An indexed sequence of bits in a buffer is allocated for tracking a battery charging state. The indexed sequence of bits has a first number of bits. A battery voltage of a rechargeable battery is sampled at a sampling rate. For each sampled battery voltage, the battery voltage is compared with a voltage threshold. A next bit position in the indexed sequence of bits is identified. In accordance with a determination that a comparison result is true, a predefined first value is added to the next bit position. A second number of bits that are filled with the predefined first value is determined. A ratio between the second number and the first number is also determined. In accordance with a determination that the ratio exceeds a threshold step-down ratio, a battery charge voltage is stepped down. The rechargeable battery is charged to a step-down voltage.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer-readable media, for speech recognition using multi-dialect and multilingual models. In some implementations, audio data indicating audio characteristics of an utterance is received. Input features determined based on the audio data are provided to a speech recognition model that has been trained to output score indicating the likelihood of linguistic units for each of multiple different language or dialects. The speech recognition model can be one that has been trained using cluster adaptive training. Output that the speech recognition model generated in response to receiving the input features determined based on the audio data is received. A transcription of the utterance generated based on the output of the speech recognition model is provided.

Abstract: The timing for displaying an overlay a video is optimized using a system that includes an engagement component that receives information regarding frames of a video at which respective users of a plurality of users stop engaging with the video and an analysis component that calculates user engagement scores for the frames based on the information, wherein the user engagement scores reflect a percentage of the plurality of users that stop engaging with the video at the respective frames, and wherein the higher the percentage the lower the user engagement score. The system further includes an identification component that identifies a subset of the frames associated with relatively lower user engagement scores in comparison to other frames of the video and an overlay component that configures an overlay for displaying over the video near at least one of the frames included in the subset.

Abstract: A method (700) of biasing a tunable laser (310) during burst-on and burst-off states includes receiving a burst mode signal (514) indicative of the burst-on state or the burst-off state and when the burst mode signal is indicative of the burst-on state: delivering a first bias current (IGAIN) to an anode of a gain-section diode (590a) disposed on a shared substrate of the tunable laser; and delivering a second bias current (IPH) to an anode of phase-section diode (590b) disposed on the shared substrate. The second bias current is less than the first bias current. When the burst mode signal transitions to be indicative of the burst-off state, the method also includes delivering the first bias current to the anode of the gain-section diode; and delivering the second bias current to the anode of the phase-section diode wherein the first bias current is less than the second bias current.

Abstract: A computing device is described for performing local interference avoidance, when supporting concurrent voice and data transmissions, and with access to multiple radios. The computing device predicts when coexistence issues will occur from maintaining independent voice and data transmissions using separate radios. To avoid local interference issues, the computing device automatically switches to operating a different combination of radios, making local interference less likely to occur. In some cases, the computing device may consolidate the voice and non-voice data exchanges to occur using a single radio. In some cases, rather than consolidation, the computing device may move the voice or the non-voice data exchange to a different radio as a way to avoid the local interference.

Abstract: In a distributed database, a transaction is to be committed at a first coordinator server and one or more participant servers 1210. The first coordinator server is configured to receive a notification that each participant server of the transaction is prepared at a respective prepared timestamp, the respective prepared timestamp being chosen within a time range for which the respective participant server obtained at least one lock 1220. The first coordinator server computes the commit timestamp for the transaction equal or greater than each of the prepared timestamps 1230, and restrict the commit timestamp such that a second coordinator server sharing at least one of the participant servers for one or more other transactions at a shared shard cannot select the same commit timestamp for any of the other transactions 1240. The transaction is committed at the commit timestamp 1250.

Abstract: Methods, systems, and media for presenting notifications indicating recommended content are provided. A notification of recommended content can be provided. An indication that a user device has initiated a casting session with a display device can be received. A request for recommended content to be presented on the display device can be received. A media content item can be identified based on at least one media content item that has been previously selected by a user account associated with the user device. A notification can be generated that includes an indication of the identified media content item and a selectable input that, when selected, causes the identified media content item to begin being presented on the display device.

Abstract: A method and apparatus for distribution of applications amongst a number of servers, ensuring that changes to application data on a master for that application are asynchronously replicated to a number of slaves for that application. Servers may be located in geographically diverse locations; the invention permits data replication over high-latency and lossy network connections and failure-tolerance under hardware and network failure conditions. Access to applications is mediated by a distributed protocol handler which allows any request for any application to be addressed to any server, and which, when working in tandem with the replication system, pauses connections momentarily to allow seamless, consistent live-migration of applications and their state between servers.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing inputs using recurrent neural networks. One of the systems includes a main recurrent neural network comprising one or more recurrent neural network layers and a respective hyper recurrent neural network corresponding to each of the one or more recurrent neural network layers, wherein each hyper recurrent neural network is configured to, at each of a plurality of time steps: process the layer input at the time step to the corresponding recurrent neural network layer, the current layer hidden state of the corresponding recurrent neural network layer, and a current hypernetwork hidden state of the hyper recurrent neural network to generate an updated hypernetwork hidden state.