Abstract: Systems and methods for reducing usage of computational resources when accessing siloed data across disparate locations via a unified metadata graph are disclosed. The system receives a user-specified query indicating a request to access a set of data objects. The system then performs natural language processing on the user-specified query to determine a set of phrases corresponding to the user-specified query. The system then accesses a metadata graph to determine a node corresponding to the set of phrases. Using a location identifier corresponding to the determined node, the system determines a data silo storing at least one data object of the set of data objects. The system then generates for display, on a graphical user interface, a visual representation of the at least one data object.

Abstract: Implementations disclosed include methods and systems that include obtaining, by a processing device, a first output from a machine learning model, wherein the first output comprises one or more data samples; identifying a first set of data samples of the one or more data samples that satisfies a threshold criterion; generating, using an explainability tool, a weighted value for each data sample of the first set of data samples; and modifying the machine learning model based at least in part on the weighted value for each data sample of the first set of data samples.

Abstract: A system includes a memory and a processing device, operatively coupled to the memory, to receive, from a client device via a user interface, input data comprising an initial version of a machine learning model, initialize an operating mode of the user interface for machine learning model building, and generate an enhanced version of the machine learning model in accordance with the operating mode.

Abstract: Systems and methods for performing data analytics on sensitive data, such as deduplication are disclosed herein. A first request to perform a first data analytics operation for a set of sensitive data instances may be received. A first set of image representations for the set of sensitive data instances may be retrieved. A first sensitive data instance and a second sensitive data instance may be clustered into a first cluster. The first data analytics operation may be performed on the first cluster.

Abstract: Systems, methods, and devices detect audio signals. Methods may include receiving an audio input at an audio front end circuit, starting, using a low power circuit, one or more buffers in response to receiving the audio input, and identifying, using the low power circuit, speech included in the audio input. Methods may also include identifying, using the low power circuit and a high performance circuit, a wake word based, at least in part, on the identified speech, the high performance circuit being configured to verify the wake word identified by the low power circuit.

Abstract: Systems, methods, and devices detect audio signals. Methods may include receiving an audio input at an audio front end circuit, and identifying, using a first circuit, a wake word based, at least in part, on the received audio signal. Methods may also include determining if the identifying performed by the first circuit should be modified based, at least in part, on a designated threshold of accuracy associated with the first circuit and a result of the identifying, the designated threshold of accuracy being determined based, at least in part, on a language processing model of a second circuit. Moreover, the modifying may include adjusting wake word detection operations of the first circuit based, at least in part, on a result of the determining.

Abstract: Systems, methods, and devices detect wake signals included in audio signals. Methods include receiving a dataset including raw audio data, the raw audio data comprising a plurality of audio samples and associated metadata, and generating, using one or more processing elements, an augmented dataset based on the raw audio data, the augmented dataset comprising a plurality of annotations identifying types of raw audio data. Methods further include generating, using the one or more processing elements, a feature dataset by extracting features from the augmented dataset based, at least in part, on the plurality of annotations, and generating, using the one or more processing elements, a wake signal detection model based, at least in part, on the feature dataset, the wake signal detection model being a machine learning model trained based on the feature dataset.

Abstract: This disclosure describes a conferencing device with beamforming and echo cancellation that includes: a microphone array that further comprises a plurality of microphones oriented to develop a corresponding plurality of microphone signals; a processor configured to execute the following steps: (1) performing a beamforming operation; (2) performing an acoustic echo cancellation operation; (3) post processing; (4) selecting one or more of the post processed signals for transmission to the far end where a signal selector uses the far end signal as information to inhibit the signal selector from changing the selection of the post processed signals while only the far end signal is active.

Abstract: This disclosure describes a conferencing device with beamforming and echo cancellation that includes: a microphone array that further comprises a plurality of microphones oriented to develop a corresponding plurality of microphone signals; a processor configured to execute the following steps: (1) performing a beamforming operation to combine the plurality of microphone signals from the microphone array into a plurality of combined signals, (2) performing an acoustic echo cancellation operation on the plurality of combined signals to generate a plurality of combined echo cancelled signals, (3) receiving with a voice activity detector the far end signal as an input, (4) selecting one or more of the combined echo cancelled signals for transmission to the far end where a signal selector uses the far end signal as information to inhibit the signal selector from changing the selection of the combined echo cancelled signals while only the far end signal is active.

Abstract: Systems and methods for generating a platform-agnostic data pipeline via a low code transformation layer are disclosed. The system receives one or more user selections of (i) nodes and (ii) links linking the nodes, indicating a data pipeline architecture of transfer/management/flow of data via a GUI. In response to receiving a user selection to implement the data pipeline, the system automatically identifies/generates a set of code portions, based on one or more software objects (e.g., JSON objects) associated with the user selections indicating the data pipeline architecture. The system then identifies a platform identifier associated with a remote server and generates a set of executable instructions (e.g., a script, executable program, or other file) associated with the data pipeline architecture by using a transformation component. The system then provides the executable instructions to the remote server to host the data pipeline.

Abstract: Described herein are devices, methods, and systems for detecting a phrase from uttered speech. A processing device may determine a first model for phrase recognition based on a likelihood ratio using a set of training utterances. The set of utterances may be analyzed by the first model to determine a second model, the second model comprising a training state sequence for each of the set of training utterances, and wherein each training state sequence indicates a likely state for each time interval of a corresponding training utterance. A determination of whether a detected utterance corresponds to the phrase may be based on a concatenation of the first model and the second model.

Abstract: A system includes memory and a processing device, operatively coupled to the memory, to obtain an input signal corresponding to data obtained from a data source, extract a set of features using the input signal, generate a set of feature tracking data from the set of features, compress a machine learning model to obtain a compressed model by identifying a subset of features based on the set of tracking data, and use the compressed model to make a prediction based on the set of feature tracking data. The set of features includes a set of confidence features and a set of uncertainty features, and the set of feature tracking data includes a set of confidence feature tracking data and a set of uncertainty feature tracking data.

Abstract: Systems and methods for performing data analytics on sensitive data, such as deduplication are disclosed herein. A first request to perform a first data analytics operation for a set of sensitive data instances may be received. A first set of image representations for the set of sensitive data instances may be retrieved. A first sensitive data instance and a second sensitive data instance may be clustered into a first cluster. The first data analytics operation may be performed on the first cluster.

Abstract: Systems and methods for generating a platform-agnostic data pipeline via a low code transformation layer are disclosed. The system receives one or more user selections of (i) nodes and (ii) links linking the nodes, indicating a data pipeline architecture of transfer/management/flow of data via a GUI. In response to receiving a user selection to implement the data pipeline, the system automatically identifies/generates a set of code portions, based on one or more software objects (e.g., JSON objects) associated with the user selections indicating the data pipeline architecture. The system then identifies a platform identifier associated with a remote server and generates a set of executable instructions (e.g., a script, executable program, or other file) associated with the data pipeline architecture by using a transformation component. The system then provides the executable instructions to the remote server to host the data pipeline.

Abstract: This disclosure describes a conferencing device with beamforming and echo cancellation that includes: a microphone array that further comprises a plurality of microphones oriented to develop a corresponding plurality of microphone signals; a processor configured to execute the following steps: (1) performing a beamforming operation to combine the plurality of microphone signals from the microphone array into a plurality of combined signals, (2) performing an acoustic echo cancellation operation on the plurality of combined signals to generate a plurality of combined echo cancelled signals, (3) receiving with a voice activity detector the far end signal as an input, (4) selecting one or more of the combined echo cancelled signals for transmission to the far end where a signal selector uses the far end signal as information to inhibit the signal selector from changing the selection of the combined echo cancelled signals while only the far end signal is active.

Abstract: Described are techniques for noise-robust and speaker-independent keyword spotting (KWS) in an input audio signal that contains keywords used to activate voice-based human-computer interactions. A KWS system may combine the latent representation generated by a denoising autoencoder (DAE) with audio features extracted from the audio signal using a machine learning approach. The DAE may be a discriminative DAE trained with a quadruplet loss metric learning approach to create a highly-separable latent representation of the audio signal in the audio input feature space. In one aspect, spectral characteristics of the audio signal such as Log-Mel features are combined with the latent representation generated by a quadruplet loss variational DAE (QVDQE) as input to a DNN KWS classifier.

Abstract: Implementations disclosed describe methods, devices, and systems to perform out-of-distribution and recognition of activities using an inertial measurement unit (IMU) sensor. A method may include receiving motion data by a device from a motion sensor. The method further includes generating image data comprising one or more images based on the motion data. The method further includes determining that a first portion of the image data corresponds to activities outside a classification distribution. The method further includes filtering the image data by removing the first portion from the image and generating filtered image data. The method further includes determining an activity classification, within the classification distribution, based on the filtered image data. The method further includes modifying an operating parameter of the device based on the activity classification.

Abstract: Implementations disclosed describe methods and systems to perform the methods of optimizing a size of memory used for accumulation of neural node outputs and for supporting multiple computational paths in neural networks. In one example, a size of memory used to perform neural layer computations is reduced by performing nodal computations in multiple batches, followed by rescaling and accumulation of nodal outputs. In another example, execution of parallel branches of neural node computations include evaluating, prior to the actual execution, the amount of memory resources needed to execute a particular order of branches sequentially and select the order that minimizes this amount or keeps this amount below a target threshold.

Abstract: Implementations disclosed describe methods and systems to perform the methods of optimizing a size of memory used for accumulation of neural node outputs and for supporting multiple computational paths in neural networks. In one example, a size of memory used to perform neural layer computations is reduced by performing nodal computations in multiple batches, followed by rescaling and accumulation of nodal outputs. In another example, execution of parallel branches of neural node computations include evaluating, prior to the actual execution, the amount of memory resources needed to execute a particular order of branches sequentially and select the order that minimizes this amount or keeps this amount below a target threshold.

Abstract: One or more computing devices, systems, and/or methods are provided. In an example, a method comprises receiving, by a device, incoming motion data from a motion sensor, generating, by the device, an incoming embedding vector based on the incoming motion data, generating, by the device, a predicted embedding vector based on the incoming embedding vector, assigning, by the device, an activity classification based on the predicted embedding vector, and modifying an operating parameter of the device based on the activity classification.