Abstract: This document relates to employing tongue gestures to control a computing device, and training machine learning models to detect tongue gestures. One example relates to a method or technique that can include receiving one or more motion signals from an inertial sensor. The method or technique can also include detecting a tongue gesture based at least on the one or more motion signals, and outputting the tongue gesture.

Abstract: A computer device is provided that includes a display device, and a sensor system configured to be mounted adjacent to a user's head and to measure an electrical potential near one or more electrodes of the sensor system. The computer device further includes a processor configured to present a periodic motion-based visual stimulus having a changing motion that is frequency-modulated for a target frequency or code-modulated for a target code, detect changes in the electrical potential via the one or more electrodes, identify a corresponding visual evoked potential feature in the detected changes in electrical potential that corresponds to the periodic motion-based visual stimulus, and recognize a user input to the computing device based on identifying the corresponding visual evoked potential feature.

Abstract: A semi-dry electrode combines advantages of wet electrodes and dry electrodes by use of a rotatable ball to apply a conductive gel at the tip of the electrode in a manner similar to how a ballpen applies ink. A reservoir in the semi-dry electrode contains the conductive gel that is applied by the ball to the skin of the user. This creates a thin film of conductive gel at the tip of the semi-dry electrode which reduces impedance and increases the signal-to-noise (SNR) ratio. Directly applying the conductive gel from within the electrode itself reduces mess and improves user convenience. The semi-dry electrode may be used in a lightweight electroencephalography (EEG) monitoring device to detect brain activity. The brain activity may be used as input for a brain-computer interface (BCI).

Abstract: A force-controlled electroencephalogram (EEG) monitoring device maintains a constant pressure between electrodes and the scalp of a user thereby increasing user comfort. Arms on the EEG monitoring device position the electrodes in contact with specific regions on the head of the user. The dimension, shape, and curvature of the arms affect the amount of force with which an electrode is held in contact with the user's scalp. The amount of pressure may be different for different regions of the user's head to achieve a balance between comfort and conductivity. The amount of pressure may be further modulated by the use of spring-loaded electrode holders that allow an electrode to move relative to the holder. To further improve user comfort, the tips of the electrodes may be hemispherical rather than pointed. The EEG monitoring device can be used as input for a brain-computer interface (BCI).

Abstract: Systems, methods, and computer-readable storage devices are disclosed for improved real-time audio processing. One method including: constructing a deep neural network model, including a plurality of at least one-bit neurons, configured to output a predicted label of audio data, the plurality of at least one-bit neurons arranged in a plurality of layers, including at least one hidden layer, and being connected by a plurality of connections, each connection having at least a one-bit weight, wherein one or both of the plurality of at least one-bit neurons and the plurality of connections have a reduced bit precision; receiving a training data set, the training data set including audio data; training the deep neural network model using the training data set; and outputting a trained deep neural network model configured to output a predicted label of real-time audio data.

Abstract: A computer device is provided that includes a display device, and a sensor system configured to be mounted adjacent to a user's head and to measure an electrical potential near one or more electrodes of the sensor system. The computer device further includes a processor configured to present a periodic motion-based visual stimulus having a changing motion that is frequency-modulated for a target frequency or code-modulated for a target code, detect changes in the electrical potential via the one or more electrodes, identify a corresponding visual evoked potential feature in the detected changes in electrical potential that corresponds to the periodic motion-based visual stimulus, and recognize a user input to the computing device based on identifying the corresponding visual evoked potential feature.

Abstract: A computer device is provided that includes a display device, and a sensor system configured to be mounted adjacent to a user's head and to measure an electrical potential near one or more electrodes of the sensor system. The computer device further includes a processor configured to present a periodic motion-based visual stimulus having a changing motion that is frequency-modulated for a target frequency or code-modulated for a target code, detect changes in the electrical potential via the one or more electrodes, identify a corresponding visual evoked potential feature in the detected changes in electrical potential that corresponds to the periodic motion-based visual stimulus, and recognize a user input to the computing device based on identifying the corresponding visual evoked potential feature.

Abstract: Systems, methods, and computer-readable storage devices are disclosed for improved audio feature discovery using a neural network. One method including: receiving a trained neural network model, the trained neural network configured to output an audio feature classification of audio data; deconstructing the trained neural network model to generate at least one saliency map, the at least one saliency map providing a successful classification of the audio feature; and extracting at least one visualization of the audio feature the trained neural network model relies on for classification based on the at least one saliency map.

Abstract: A computer device is provided is includes a display device, and a sensor system configured to be mounted adjacent to a user's head and to measure an electrical potential near one or more electrodes of the sensor system. The computer device further includes a processor configured to present a periodic motion-based visual stimulus having a changing motion that is frequency-modulated for a target frequency or code-modulated for a target code, detect changes in the electrical potential via the one or more electrodes, identify a corresponding visual evoked potential feature in the detected changes in electrical potential that corresponds to the periodic motion-based visual stimulus, and recognize a user input to the computing device based on identifying the corresponding visual evoked potential feature.

Abstract: A computing system that facilitates decoding a spherical harmonics (SH) representation of a three-dimensional sound signal to a binaural sound signal is described herein. The computing system generates a binaural sound signal based upon the SH representation, a tapering window function that is selected based on an SH encoding order of the SH representation, and a coloration compensation filter that incorporates the tapering window function. The computing system causes the binaural sound signal to be played over at least two speakers.

Abstract: A facility for processing output from a network of mechanical sensors is described. The facility accesses time-series data outputted by the network of sensors. The facility applies to the accessed time-series data a trained autoencoder to obtain a version of the accessed time-series data in which noise present in the accessed time-series data is at least partially suppressed. The facility stores the obtained version of the accessed time-series data, such as in order to perform human activity recognition against the obtained version of the accessed time-series data.

Abstract: A computing system that facilitates decoding a spherical harmonics (SH) representation of a three-dimensional sound signal to a binaural sound signal is described herein. The computing system generates a binaural sound signal based upon the SH representation, a tapering window function that is selected based on an SH encoding order of the SH representation, and a coloration compensation filter that incorporates the tapering window function. The computing system causes the binaural sound signal to be played over at least two speakers.

Abstract: Systems, methods, and computer-readable storage devices are disclosed for improved real-time audio processing. One method including: receiving audio data including a plurality of frames having a plurality of frequency bins; calculating, for each frequency bin, an approximate speech signal estimation based on the plurality of frames; calculating, for each approximate speech signal estimation, a clean speech estimation and at least one additional target including an ideal ratio mask using a trained neural network model; and calculating, for each frequency bin, a final clean speech estimation using the calculated at least one additional target including the calculated ideal ratio mask and the calculated clean speech estimation.

Abstract: Systems, methods, and computer-readable storage devices are disclosed for improved audio feature discovery using a neural network. One method including: receiving a trained neural network model, the trained neural network configured to output an audio feature classification of audio data; deconstructing the trained neural network model to generate at least one saliency map, the at least one saliency map providing a successful classification of the audio feature; and extracting at least one visualization of the audio feature the trained neural network model relies on for classification based on the at least one saliency map.

Abstract: An ultrasonic gesture recognition system is provided that recognizes gestures based on analysis of return signals of an ultrasonic pulse that is reflected from a gesture. The system transmits an ultrasonic chirp and samples a microphone array at sample intervals to collect a return signal for each microphone. The system then applies a beamforming technique to frequency domain representations of the return signals to generate an acoustic image with a beamformed return signal for multiple directions. The system then generates a feature image from the acoustic images to identify, for example, distance or depth from the microphone array to the gesture for each direction. The system then submits the feature image to a deep learning system to classify the gesture.

Abstract: Systems, methods, and computer-readable storage devices are disclosed for improved real-time audio processing. One method including: constructing a deep neural network model, including a plurality of at least one-bit neurons, configured to output a predicted label of audio data, the plurality of at least one-bit neurons arranged in a plurality of layers, including at least one hidden layer, and being connected by a plurality of connections, each connection having at least a one-bit weight, wherein one or both of the plurality of at least one-bit neurons and the plurality of connections have a reduced bit precision; receiving a training data set, the training data set including audio data; training the deep neural network model using the training data set; and outputting a trained deep neural network model configured to output a predicted label of real-time audio data.

Abstract: Systems, methods, and computer-readable storage devices are disclosed for improved real-time audio processing. One method including: receiving audio data including a plurality of frames having a plurality of frequency bins; calculating, for each frequency bin, an approximate speech signal estimation based on the plurality of frames; calculating, for each approximate speech signal estimation, a clean speech estimation and at least one additional target including an ideal ratio mask using a trained neural network model; and calculating, for each frequency bin, a final clean speech estimation using the calculated at least one additional target including the calculated ideal ratio mask and the calculated clean speech estimation.

Abstract: A system is provided that employs a statistical approach to semi-supervised speech enhancement with a low-order non-negative matrix factorization (“NMF”). The system enhances noisy speech based on multiple dictionaries with dictionary atoms derived from the same clean speech samples and generates an enhanced speech representation of the noisy speech by combining, for each dictionary, a clean speech representation of the noisy speech generated based on a NMF using the dictionary atoms of the dictionary. The system generates frequency-domain (“FD”) clean speech sample representations of the clean speech samples, for example, using a Fourier transform. To generate each dictionary, the system generates a dictionary-unique initialization of the dictionary atoms and the activations and performs a NMF of the FD clean speech samples.

Abstract: Systems and methods of providing an audio signal are disclosed herein. In one embodiment, a method of delivering an audio signal from a device toward a user's ear includes, for example, transmitting a filtered audio signal from a transducer positioned at a location on the device that is longitudinally spaced apart from an entrance of an auditory canal of the user's ear when the device is worn on the user's head.

Abstract: A facility for processing output from a network of mechanical sensors is described. The facility accesses time-series data outputted by the network of sensors. The facility applies to the accessed time-series data a trained autoencoder to obtain a version of the accessed time-series data in which noise present in the accessed time-series data is at least partially suppressed. The facility stores the obtained version of the accessed time-series data, such as in order to perform human activity recognition against the obtained version of the accessed time-series data.