Abstract: Implementations described herein identify and correct automatic speech recognition (ASR) misrecognitions. For example, on-device processor(s) of a client device may generate a predicted textual segment that is predicted to correspond to spoken utterance of a user of the client device, and may receive further input that modifies the predicted textual segment to an alternate textual segment. Further, the on-device processor(s) may store these textual segments in on-device storage as a candidate correction pair, and transmit the candidate correction pair to a remote system. Moreover, remote processor(s) of the remote system may determine that the candidate correction pair is an actual correction pair, and may cause client devices to generate updates for a global ASR model for the candidate correction pair. Additionally, the remote processor(s) may distribute the global ASR model to the client devices and/or additional client devices.

Abstract: Systems and methods for compression of data that exhibits mixed compressibility, such as floating-point data, are provided. As one example, aspects of the present disclosure can be used to compress floating-point data that represents the values of parameters of a machine-learned model. Therefore, aspects of the present disclosure can be used to compress machine-learned models (e.g., for reducing storage requirements associated with the model, reducing the bandwidth expended to transmit the model, etc.).

Abstract: Processor(s) of a client device can: receive audio data that captures a spoken utterance of a user of the client device; process, using an on-device speech recognition model, the audio data to generate a predicted textual segment that is a prediction of the spoken utterance; cause at least part of the predicted textual segment to be rendered (e.g., visually and/or audibly); receive further user interface input that is a correction of the predicted textual segment to an alternate textual segment; and generate a gradient based on comparing at least part of the predicted output to ground truth output that corresponds to the alternate textual segment. The gradient is used, by processor(s) of the client device, to update weights of the on-device speech recognition model and/or is transmitted to a remote system for use in remote updating of global weights of a global speech recognition model.

Abstract: Systems and methods for compression of data that exhibits mixed compressibility, such as floating-point data, are provided. As one example, aspects of the present disclosure can be used to compress floating-point data that represents the values of parameters of a machine-learned model. Therefore, aspects of the present disclosure can be used to compress machine-learned models (e.g., for reducing storage requirements associated with the model, reducing the bandwidth expended to transmit the model, etc.).

Abstract: Processor(s) of a client device can: receive audio data that captures a spoken utterance of a user of the client device; process, using an on-device speech recognition model, the audio data to generate a predicted textual segment that is a prediction of the spoken utterance; cause at least part of the predicted textual segment to be rendered (e.g., visually and/or audibly); receive further user interface input that is a correction of the predicted textual segment to an alternate textual segment; and generate a gradient based on comparing at least part of the predicted output to ground truth output that corresponds to the alternate textual segment. The gradient is used, by processor(s) of the client device, to update weights of the on-device speech recognition model and/or is transmitted to a remote system for use in remote updating of global weights of a global speech recognition model.

Abstract: Implementations disclosed herein are directed to efficient federated learning of machine learning (ML) model(s) at a remote system (e.g., remote server(s)) based on update(s) generated at client device(s). Processor(s) of the client device(s) can receive client data, process, using on-device ML model(s), the client data to generate predicted output(s), generate, using unsupervised learning, gradient(s) based on the predicted output(s), generate, based on the gradient(s), the update(s) for disparate portions of the on-device ML model(s) and/or global ML model(s) that are remote-based counterparts of the on-device ML model(s). Further, processor(s) of the remote system can receive, from the client device(s), the update(s) for the disparate portions of the on-device ML model(s), and cause the global ML model(s) to be updated based on the update(s) for the disparate portions of the on-device ML model(s) received from disparate client device(s).

Abstract: A computer-implemented method includes obtaining a multi-domain (MD) dataset and training a neural network model using the MD dataset with short-form data withheld (MD-SF). The neural network model includes a plurality of layer each having a plurality of parameters. The method also includes resetting each respective layer in the trained neural network one at a time. For each respective layer in the trained neural network model, and after resetting the respective layer, the method also includes determining a corresponding word error rate of the trained neural network model and identifying the respective layer as corresponding to an ambient layer when the corresponding word error rate satisfies a word error rate threshold. The method also includes transmitting an on-device neural network model to execute on one or more client devices for generating gradients based on the withheld domain (SF) of the MD dataset.

Abstract: Systems and methods for compression of data that exhibits mixed compressibility, such as floating-point data, are provided. As one example, aspects of the present disclosure can be used to compress floating-point data that represents the values of parameters of a machine-learned model. Therefore, aspects of the present disclosure can be used to compress machine-learned models (e.g., for reducing storage requirements associated with the model, reducing the bandwidth expended to transmit the model, etc.).

Abstract: Processor(s) of a client device can: receive audio data that captures a spoken utterance of a user of the client device; process, using an on-device speech recognition model, the audio data to generate a predicted textual segment that is a prediction of the If spoken utterance; cause at least part of the predicted textual segment to be rendered (e.g., visually and/or audibly); receive further user interface input that is a correction of the predicted textual segment to an alternate textual segment; and generate a gradient based on comparing at least part of the predicted output to ground truth output that corresponds to the alternate textual segment. The gradient is used, by processor(s) of the client device, to update weights of the on-device speech recognition model and/or is transmitted to a remote system for use in remote updating of global weights of a global speech recognition model.

Abstract: A computer-implemented method performed by a data processing apparatus includes receiving an audio signal that includes a frequency-domain representation of an audio file, extracting, from the audio signal, a plurality of frequency-domain data values that correspond to at least a portion of the audio file, compressing the plurality of data values to form a compressed frequency domain value file, and transmitting the compressed frequency domain value file to a server to identify the audio file.

Abstract: Systems and techniques are provided for finding differences in nearly-identical audio recordings. A first version of an audio recording may be received. A second version of the audio recording may be received. A difference between the first version of the audio recording and the second version of the audio recording may be determined using time domain analysis and frequency domain analysis. The difference may be stored in a difference set. The difference set may allow the first version of the audio recording to be distinguished from the second version of the audio recording. The audio recording may be a music track. The first version of the audio recording may be an explicit version of the music track. The second version of the audio recording may be an edited version of the music track.

Abstract: A computer-implemented method performed by a data processing apparatus includes receiving an audio signal that includes a frequency-domain representation of an audio file, extracting, from the audio signal, a plurality of frequency-domain data values that correspond to at least a portion of the audio file, compressing the plurality of data values to form a compressed frequency domain value file, and transmitting the compressed frequency domain value file to a server to identify the audio file.

Abstract: A method for updating a computer the includes converting a first computer the to executable byte code and receiving a second computer the that includes a change that distinguishes the second computer file from the first computer file. The method also includes converting the second computer the to executable byte code and comparing at least a portion of the executable byte code of the second computer file with at least a portion of the executable byte code of the first computer file. The method further includes inserting, into the executable byte code the first computer file, a modification that causes at least a portion of the executable byte code of the first computer the to resemble the executable byte code of the second computer file.

Abstract: Systems and techniques are provided for finding differences in nearly-identical audio recordings. A first version of an audio recording may be received. A second version of the audio recording may be received. A difference between the first version of the audio recording and the second version of the audio recording may be determined using time domain analysis and frequency domain analysis. The difference may be stored in a difference set. The difference set may allow the first version of the audio recording to be distinguished from the second version of the audio recording. The audio recording may be a music track. The first version of the audio recording may be an explicit version of the music track. The second version of the audio recording may be an edited version of the music track.

Abstract: This disclosure describes techniques associated with filtering of video data in a video encoding and/or decoding process. In accordance with this disclosure, filtering is applied at an encoder, and filter information is encoded in the bitstream to identify the filtering that was applied at the encoder. Different types of filtering may be applied based on an activity metric determined for the video data. Moreover, in accordance with this disclosure, the manner in which the filter information is encoded into the bitstream may be dependent on the activity metric. In particular, for a first range of the activity metric, one or more filters are encoded directly, and for a second range of the activity metric, one or more filters are predictively encoded.

Abstract: A method, system, and device supporting installation of updates to software and/or firmware in memory of an electronic device is described. The update information may be automatically generated based upon a list of all software components or packages already present on the electronic device, and may comprise software components for all dependencies of an application being installed by the update that are not already present on the electronic device.

Abstract: A method, system, and device supporting installation of updates to software and/or firmware in memory of an electronic device is described. The update information may be automatically generated based upon a list of all software components or packages already present on the electronic device, and may comprise software components for all dependencies of an application being installed by the update that are not already present on the electronic device.

Abstract: This disclosure describes techniques for encoding digital video data using interpolation filters and offsets. An encoder may be configured to select interpolation filters for sub-pixel precision motion estimation based on historical interpolation results obtained for previously encoded video units, such as frames or slices. The encoder also may be configured to compute and assign offsets to the sub-pixel positions after interpolation based on differences between a reference unit and the unit to be coded. The computation and assignment of offsets may be performed before motion estimation. Motion estimation may be refined so that the motion search considers sub-pixel positions to which offsets have been previously added and evaluates sub-pixel positions that have a non-zero offset. In some cases, interpolation filter selection, offset computation, and/or refined motion estimation for a given unit may be performed in a single encoding pass.

Abstract: Systems and methods to process motion vectors of video data are disclosed. According to an embodiment, an intra-block in a video frame of the video data is identified. At least a first set and a second set of inter-blocks that neighbor the identified intra-block are identified, where each inter-block in a set of inter-blocks has a motion vector associated therewith. Data of the first and second set of inter-blocks is evaluated to determine an error associated with each of the first and second sets. A motion vector associated with motion vectors of the inter-blocks of the set with the least error is determined. The determined motion vector may be associated with the identified intra-block.

Abstract: A method, system, and device supporting installation of updates to software and/or firmware in memory of an electronic device is described. The update information may be automatically generated based upon a list of all software components or packages already present on the electronic device, and may comprise software components for all dependencies of an application being installed by the update that are not already present on the electronic device.