Multilayered determination of health events using resource-constrained platforms
Detecting and identifying a predetermined health event can include detecting a potential occurrence of the predetermined health event for a user by processing in real-time motion signals corresponding to motion of the user. A likelihood that the potential occurrence is an actual occurrence of the predetermined health event can be determined based on template matching of the motion signals. In response to determining that the likelihood exceeds a predetermined threshold, audio signals coinciding in time with the motion of the user can be processed using one or more layers of a multilayered audio event classifier.
1. A method, comprising:
detecting, using a user device, a potential occurrence of a predetermined health event for a user by processing in real-time motion signals corresponding to motion of the user;
determining a likelihood that the potential occurrence is an actual occurrence of the predetermined health event based on template matching of the motion signals; and
in response to determining that the likelihood exceeds a predetermined threshold, processing audio signals coinciding in time with the motion of the user using one or more layers of a multilayered audio event classifier.
2. The method of claim 1 , wherein
the template matching is performed using a self-tuning multi-centroid classifier.
3. The method of claim 1 , wherein
the processing the audio signals includes:
in response to detecting, at a first layer of the multilayered audio event classifier, audio signals corresponding to voice activity, filtering the audio signals to separate the audio signals corresponding to voice activity from audio signals corresponding to non-voice activity;
in response to detecting, at a second layer of the multilayered audio event classifier, one or more sharp sound events within the audio signals corresponding to voice activity, processing the audio signals corresponding to voice activity to separate the one or more sharp sound events contained therein from non-sharp sound events; and
classifying the sharp sound events using a machine learning classifier.
4. The method of claim 3 , further comprising:
selecting one of a plurality of machine learning classifiers for classifying the sharp sound events, wherein the selecting is based on determining an availability of resources of the device.
5. The method of claim 1 , further comprising:
conveying the audio signals to an auxiliary device communicatively coupled with the device, wherein the processing the audio signals with a multilayered audio event classifier is performed by the auxiliary device.
6. The method of claim 5 , wherein
the conveying the audio signals is performed in response to detecting that a power level of the device is less than a predetermined threshold.
7. The method of claim 5 , wherein
the device is a wearable device, and the auxiliary device comprises a plurality of auxiliary devices, and wherein the method further comprises:
detecting that the wearable device is positioned to impede generating at least one of the motion signals or the audio signals using one or more sensors of the wearable device; and
automatically selecting one of the plurality of auxiliary devices for generating the at least one of the motion signals or the audio signals.
8. The method of claim 1 , further comprising:
detecting at least one of acoustic noise or motion noise above a predetermined threshold within a selected segment of at least one of the motion signals or the audio signals; and
discarding the selected segment.
9. The method of claim 1 , wherein
the actual health event is a pulmonary event, and wherein the method further comprises:
performing a lung function parameter estimation in response to the pulmonary event.
10. The method of claim 9 , further comprising:
performing an orientation calibration that includes at least one of generating a user-specific template for retraining a model used for the template matching or generating a transfer function based on a user-specific baseline for calibrating a motion sensor for sensing the real-time motion signals.
11. A system, comprising:
one or more processors configured to initiate operations including:
detecting a potential occurrence of a predetermined health event for a user by processing in real-time motion signals corresponding to motion of the user;
determining a likelihood that the potential occurrence is an actual occurrence of the predetermined health event based on template matching of the motion signals; and
in response to determining that the likelihood exceeds a predetermined threshold, processing audio signals coinciding in time with the motion of the user using one or more layers of a multilayered audio event classifier.
12. The system of claim 11 , wherein
the template matching is performed using a self-tuning multi-centroid classifier.
13. The system of claim 11 , wherein
the processing the audio signals includes:
in response to detecting, at a first layer of the multilayered audio event classifier, audio signals corresponding to voice activity, filtering the audio signals to separate the audio signals corresponding to voice activity from audio signals corresponding to non-voice activity;
in response to detecting, at a second layer of the multilayered audio event classifier, one or more sharp sound events within the audio signals corresponding to voice activity, processing the audio signals corresponding to voice activity to separate the one or more sharp sound events contained therein from non-sharp sound events; and
classifying the sharp sound events using a machine learning classifier.
14. The system of claim 13 , wherein the processor is configured to initiate operations further including:
selecting one of a plurality of machine learning classifiers for classifying the sharp sound events, wherein the selecting is based on determining an availability of resources of a device in which the system is implemented.
15. The system of claim 11 , wherein the processor is configured to initiate operations further including:
conveying the audio signals from a device in which the system is implemented to an auxiliary device communicatively coupled with the device, wherein the processing the audio signals with a multilayered audio event classifier is performed by the auxiliary device.
16. The system of claim 15 , wherein
the conveying the audio signals is performed in response to detecting that a power level of the device is less than a predetermined threshold.
17. The system of claim 15 , wherein
the device is a wearable device, and the auxiliary device comprises a plurality of auxiliary devices, and wherein the processor is configured to initiate operations further including:
detecting that the wearable device is positioned to impede generating at least one of the motion signals or the audio signals using one or more sensors of the wearable device; and
automatically selecting one of the plurality of auxiliary devices for generating the at least one of the motion signals or the audio signals.
18. The system of claim 11 , wherein the processor is configured to initiate operations further including:
detecting at least one of acoustic noise or motion noise above a predetermined threshold within a selected segment of at least one of the motion signals or the audio signals; and
discarding the selected segment.
19. The system of claim 11 , wherein
the actual health event is a pulmonary event, and wherein the processor is configured to initiate operations further including:
performing a lung function parameter estimation in response to the pulmonary event.
20. A computer program product, the computer program product comprising:
one or more computer-readable storage media and program instructions collectively stored on the one or more computer-readable storage media, the program instructions executable by a processor to cause the processor to initiate operations including:
detecting a potential occurrence of a predetermined health event for a user by processing in real-time motion signals corresponding to motion of the user;
determining a likelihood that the potential occurrence is an actual occurrence of the predetermined health event based on template matching of the motion signals; and
in response to determining that the likelihood exceeds a predetermined threshold, processing audio signals coinciding in time with the motion of the user using one or more layers of a multilayered audio event classifier.