Abstract: The subject technology provides physiological state prediction based on acoustic data using machine learning. An apparatus receives input data comprising acoustic signal information associated with a user. The apparatus extracts one or more acoustic features from the acoustic signal information. The apparatus produces a trained machine learning model by training a neural network to predict one or more physiological states of the user from the one or more acoustic features.

Abstract: A system can receive an input indicating a user condition. The system can also receive an internal audio stream from an in-ear microphone and an external audio stream from an external microphone of a head worn system. The system can determine a respiration rate of a user based on the internal audio stream, the external audio stream, and the input indicating the user condition. In some implementations, the respiration rate may be determined from a respiration signal in the internal audio stream and/or the external audio stream. The respiration signal may measure breathing of the user. In some implementations, the system can invoke a machine learning model to determine the respiration signal from the internal audio stream and/or the external audio stream based on the user condition.

Abstract: A head-mounted device having a plurality of electrodes configured to detect optical events such as the movement of one or more eyes or coarse eye gestures is disclosed. In some examples, the one or more electrodes can be coupled to dielectric elastomer materials whose shape can be changed to vary contact between a user of the head-mounted device and the one or more electrodes to ensure sufficient contact and electrode signal quality. In some examples, the one or more electrodes can be coupled to pressure sensors and control circuitry to monitor and adjust the applied pressure. In some examples, the optical events can be used as triggers for operating the device, including transitioning between operational power modes. In some examples, the triggers can invoke higher resolution sensing capabilities of the head-mounted device. In some examples, the electrodes can be used as an on-head detector to wake-up and/or unlock the device.

Abstract: Embodiments are disclosed for ECG signal reconstruction from an EEG signal. In some embodiments, a method comprises: determining, with at least one processor, heartbeat timestamps from a heartrate signal; obtaining an electroencephalography (EEG) signal from an EEG sensor; aligning, with the at least one processor, the EEG signal with the heartbeat timestamps; collaborative filtering, with the at least one processor, the beat-aligned EEG signal; and determining, with the at least one processor, an approximation of an electrocardiography (ECG) signal based on the filtered beat-aligned EEG signal.

Abstract: The subject technology provides frequency-aware masked autoencoders for multimodal pretraining on frequency-based signals. An apparatus receives input data comprising frequency-based signal information associated with one or more modalities. The apparatus transforms the input data from a time domain to a frequency domain. The apparatus generates a frequency-embedded latent representation of the input data comprising time-domain and frequency-domain information. The apparatus also generates a masked frequency-embedded latent representation by masking one or more frequency components in the frequency-embedded latent representation. The apparatus produces a trained machine learning model by training a neural network to predict one or more masked frequency components of the frequency-embedded latent representation.

Abstract: Electrodes that can be formed in a flexible band of a wrist-worn device to detect hand gestures are disclosed. Multiple rows of electrodes can be configured to detect electromyography (EMG) signals produced by activity of muscles and tendons. The band can include removable electrical connections (e.g., pogo pins) to enable the electrode signals to be routed to processing circuitry in the housing of the wrist-worn device. Measurements between signals from the active electrodes and one or more reference electrodes can be obtained to capture EMG signals at a number of locations on the band. The measurement method and mode of operation (lower power coarse detection or higher power fine detection) can determine the location and number of electrodes to be measured. These EMG signals can be processed to identify hand movements and recognize gestures associated with those hand movements.

Abstract: Techniques are disclosed for defining a training data set to include biosignals and categorical labels representative of a context. For example, a categorical label may indicate whether a user was performing a difficult or easy mental task while the biosignal was being recorded. A set of first layers in a neural network can be trained using a portion of the training data set associated with a first set of users and at least one second layer can be trained using a portion of the training data set associated with a particular other user. The neural network can then be used to process other biosignals from the particular other user to generate predicted categorical context labels.

Abstract: Electrodes that can be formed in a flexible band of a wrist-worn device to detect hand gestures are disclosed. Multiple rows of electrodes can be configured to detect electromyography (EMG) signals produced by activity of muscles and tendons. The band can include removable electrical connections (e.g., pogo pins) to enable the electrode signals to be routed to processing circuitry in the housing of the wrist-worn device. Measurements between signals from the active electrodes and one or more reference electrodes can be obtained to capture EMG signals at a number of locations on the band. The measurement method and mode of operation (lower power coarse detection or higher power fine detection) can determine the location and number of electrodes to be measured. These EMG signals can be processed to identify hand movements and recognize gestures associated with those hand movements.

Abstract: A head-mounted device having a plurality of electrodes configured to detect optical events such as the movement of one or more eyes or coarse eye gestures is disclosed. In some examples, the one or more electrodes can be coupled to dielectric elastomer materials whose shape can be changed to vary contact between a user of the head-mounted device and the one or more electrodes to ensure sufficient contact and electrode signal quality. In some examples, the one or more electrodes can be coupled to pressure sensors and control circuitry to monitor and adjust the applied pressure. In some examples, the optical events can be used as triggers for operating the device, including transitioning between operational power modes. In some examples, the triggers can invoke higher resolution sensing capabilities of the head-mounted device. In some examples, the electrodes can be used as an on-head detector to wake-up and/or unlock the device.

Abstract: Aspects of the subject technology relate to providing gesture-based control of electronic devices. Providing gesture-based control may include determining, with a machine learning system that includes multiple machine learning models, a prediction of one or more gestures and their corresponding probabilities of being performed. A likelihood of the user's intent to actually perform that gesture may then be generated, based on the prediction and a gesture detection factor. The likelihood may be dynamically updated over time, and a visual, auditory, and/or haptic indicator of the likelihood may be provided as user feedback. The visual, auditory, and/or haptic indicator may be helpful to guide the user to the correct gesture if the gesture is intended, or to stop performing an action similar to the gesture if the gesture is not intended.

Abstract: Methods, systems and/or computer-implemented instructions are configured to perform or support actions that include: determining a time contribution for each of a set of workout effort zones for a user, wherein each of the set of workout effort zones corresponds to a range of values for a biosignal; determining a timeseries of workout target effort zones for the user based on the target time contributions for the set of workout effort zones; receiving, during a workout time period, real-time biosignal data from a sensor in a wearable electronic device being worn by the user; generating, during the workout time period, an audio, visual, or haptic stimulus based on the real-time biosignal data and a target effort zone in the time series of workout target effort zones; and outputting, during the workout time period, the audio, visual, or haptic stimulus.

Abstract: A head-mounted device having a plurality of electrodes configured to detect optical events such as the movement of one or more eyes or coarse eye gestures is disclosed. In some examples, the one or more electrodes can be coupled to dielectric elastomer materials whose shape can be changed to vary contact between a user of the head-mounted device and the one or more electrodes to ensure sufficient contact and electrode signal quality. In some examples, the one or more electrodes can be coupled to pressure sensors and control circuitry to monitor and adjust the applied pressure. In some examples, the optical events can be used as triggers for operating the device, including transitioning between operational power modes. In some examples, the triggers can invoke higher resolution sensing capabilities of the head-mounted device. In some examples, the electrodes can be used as an on-head detector to wake-up and/or unlock the device.

Abstract: Techniques are disclosed for defining a training data set to include biosignals and categorical labels representative of a context. For example, a categorical label may indicate whether a user was performing a difficult or easy mental task while the biosignal was being recorded. A set of first layers in a neural network can be trained using a portion of the training data set associated with a first set of users and at least one second layer can be trained using a portion of the training data set associated with a particular other user. The neural network can then be used to process other biosignals from the particular other user to generate predicted categorical context labels.

Abstract: Techniques are disclosed for defining a training data set to include biosignals and categorical labels representative of a context. For example, a categorical label may indicate whether a user was performing a difficult or easy mental task while the biosignal was being recorded. A set of first layers in a neural network can be trained using a portion of the training data set associated with a first set of users and at least one second layer can be trained using a portion of the training data set associated with a particular other user. The neural network can then be used to process other biosignals from the particular other user to generate predicted categorical context labels.

Abstract: A wearable electronic device includes a housing, and an electrode carrier attached to the housing and having a nonplanar surface. The wearable electronic device includes a set of electrodes, including electrodes positioned at different locations on the nonplanar surface. The wearable electronic device includes a sensor circuit and a switching circuit. The switching circuit is operable to electrically connect a number of different subsets of one or more electrodes in the set of electrodes to the sensor circuit.

Abstract: Electrodes that can be formed in a flexible band of a wrist-worn device to detect hand gestures are disclosed. Multiple rows of electrodes can be configured to detect electromyography (EMG) signals produced by activity of muscles and tendons. The band can include removable electrical connections (e.g., pogo pins) to enable the electrode signals to be routed to processing circuitry in the housing of the wrist-worn device. Measurements between signals from the active electrodes and one or more reference electrodes can be obtained to capture EMG signals at a number of locations on the band. The measurement method and mode of operation (lower power coarse detection or higher power fine detection) can determine the location and number of electrodes to be measured. These EMG signals can be processed to identify hand movements and recognize gestures associated with those hand movements.

Abstract: A head-mounted device having a plurality of electrodes configured to detect optical events such as the movement of one or more eyes or coarse eye gestures is disclosed. In some examples, the one or more electrodes can be coupled to dielectric elastomer materials whose shape can be changed to vary contact between a user of the head-mounted device and the one or more electrodes to ensure sufficient contact and electrode signal quality. In some examples, the one or more electrodes can be coupled to pressure sensors and control circuitry to monitor and adjust the applied pressure. In some examples, the optical events can be used as triggers for operating the device, including transitioning between operational power modes. In some examples, the triggers can invoke higher resolution sensing capabilities of the head-mounted device. In some examples, the electrodes can be used as an on-head detector to wake-up and/or unlock the device.

Abstract: Aspects of the subject technology relate to gesture-control inputs to an electronic device for controlling one or more other devices. The electronic device can efficiently provide gesture control for multiple other devices by mapping a finite set of user gestures to a specific set of gesture-control elements for each of the multiple other devices. In this way a single gesture can be detected for potentially controlling various different functions of various different devices. Prior to gesture control, the electronic device may receive a selection of a particular one of the multiple other devices for control, and obtain the specific set of gesture-control elements for gesture control of that selected device.

Abstract: Methods and systems are provided for predicting cognitive load. A computing device receives sensor measurements from sensors. The sensor measurements correspond to characteristics of a user during the performance of a task. For each sensor, the computing device derives, from the sensor measurements of the sensor, a set of features predictive of the cognitive load of the user; generates, from those features, a self-attention vector that characterizes each feature of the set of features relative to another feature; and defines a feature vector from the features and the self-attention vector. The computing device generates an input feature vector from the feature vector of at least one sensor. The computing device then uses a machine-learning model to generate an indication of the cognitive load of the user during the performance of a task from the feature vector.

Abstract: The present application relates to apparatus, systems, and methods to perform subject-aware self-supervised learning of a machine-learning model for classification of data, such as classification of biosignals.