Pulmonary function estimation

1. A method within and by a computer hardware system including a hardware processor, comprising:

passively monitoring a time series of audio signals generated by a microphone of the computer hardware system for a cough event by a user;

filtering, by a cough detector within the hardware processor, portions of the audio signals that do not relate to the cough event to generate filtered audio signals;

generating, by a feature vector generator, a cough feature vector corresponding to the filtered audio signals, wherein each element of the cough feature vector corresponds to one or more of a plurality of audio signal features extracted from the filtered audio signals, wherein the plurality of audio signal features includes mean value of mel-frequency cepstral coefficients (MFCCs) and skewness of MFCCs; and

generating, by a regressor implemented by the hardware processor, a predicted lung function metric of the user based on the cough feature vector;

wherein the computer hardware system includes a feature vector optimizer configured to select particular ones of the plurality of audio signal features as extracted from the filtered audio signals for inclusion into the cough feature vector based upon one or more biodemographic factors of the user including a health status of the user.

2. The method of claim 1 , wherein

the plurality of audio signal features includes cough force.

3. The method of claim 1 , wherein

the plurality of audio signal features includes cough wheeziness and the one or more biodemographic factors include age of the user.

4. The method of claim 1 , further comprising:

determining, with a geo-positioning sensor, a geographic location of the user;

determining, with the hardware processor, an environmental condition of the geographic location; and

wherein the generating, by the regressor implemented by the hardware processor, the predicted lung function metric of the user is based on the cough feature vector and the environmental condition.

5. The method of claim 1 , wherein a context-based regression model is selected from a plurality of context-based regression models to determine the predicted lung function metric based on at least one of the one or more biodemographic factors of the user, and wherein each context-based regression model of the plurality of context-based regression models is based on a distinct group of subjects classified according to the at least one of the one or more biodemographic factors.

6. The method of claim 5 , wherein

the regressor is configured to implement a condition-relevant regression model based upon the at least one of the one or more biodemographic factors of the user.

7. The method of claim 1 , wherein

the computer hardware system is configured to:

determine a quality of the cough event, at least in part, as an amount of noise in an audio signal provided by the cough event, and

present, via the computer hardware system, an instruction to the user to perform an active lung function metric measurement responsive to the quality of the cough event having the audio signal with the amount of noise above a predetermined threshold.

8. The method of claim 1 , wherein

the cough detector includes a cough segmenter configured to extract edges of a burst phase of the cough event and reject non-burst portions of the cough event, and

the cough feature vector is generated only based upon the burst phase of the cough event.

9. The method of claim 1 , wherein

the predicted lung function metric is a pulmonary function test value including one or more of FEV1, FVC, or a FEV1/FVC ratio.

10. A computer hardware system, comprising:

a microphone configured to generate a time series of audio signals; and

a hardware processor configured to passively monitor the time series of the audio signals for a cough event by a user, wherein the hardware processor includes

a cough detector configured to filter portions of the audio signals that do not relate to the cough event to generate filtered audio signals;

a feature vector generator configured to generate a cough feature vector corresponding to the filtered audio signals, wherein each element of the cough feature vector corresponds to one or more of a plurality of audio signal features extracted from the filtered audio signals, wherein the plurality of audio signal features includes mean value of mel-frequency cepstral coefficients (MFCCs) and skewness of MFCCs; and

a regressor configured to generate a predicted lung function metric of the user based on the cough feature vector;

wherein the computer hardware system includes a feature vector optimizer configured to select particular ones of the plurality of audio signal features as extracted from the filtered audio signals for inclusion into the cough feature vector based upon one or more biodemographic factors of the user including a health status of the user.

11. The system of claim 10 , wherein

the plurality of audio signal features extracted from the filtered audio signals includes cough force.

12. The system of claim 10 , wherein

the plurality of audio signal features includes cough wheeziness and the one or more biodemographic factors include age of the user.

13. The system of claim 10 , further comprising:

a geo-positioning sensor to determine a geographic location of the user;

an environmental condition determiner configured to determine an environmental condition of the geographic location; and

wherein the regressor is configured to generate the predicted lung function metric of the user based on the cough feature vector and the geographic location.

14. The system of claim 10 , wherein a context-based regression model is selected from a plurality of context-based regression models to determine the predicted lung function metric based on at least one of the one or more biodemographic factors of the user, and wherein each context-based regression model of the plurality of context-based regression models is based on a distinct group of subjects classified according to the at least one of the one or more biodemographic factors.

15. The system of claim 14 , wherein

the regressor is configured to implement a condition-relevant regression model based upon the at least one of the one or more biodemographic factors of the user.

16. The system of claim 10 , wherein

the computer hardware system is configured to:

determine a quality of the cough event, at least in part, as an amount of noise in an audio signal provided by the cough event, and

present, via the computer hardware system, an instruction to the user to perform an active lung function metric measurement responsive to the quality of the cough event having the audio signal with the amount of noise above a predetermined threshold.

17. The system of claim 10 , wherein

the cough detector includes a cough segmenter configured to extract edges of a burst phase of the cough event and reject non-burst portions of the cough event, and

the cough feature vector is generated only based upon the burst phase of the cough event.

18. A method within and by a computer hardware system including a hardware processor, comprising:

passively monitoring a time series of audio signals generated by a microphone of the computer hardware system for a cough event by a user;

filtering, by a cough detector within the hardware processor, portions of the audio signals that do not relate to the cough event to generate filtered audio signals;

generating, by a feature vector generator, a cough feature vector corresponding to the filtered audio signals, wherein each element of the cough feature vector corresponds to an audio signal feature extracted from the filtered audio signals; and

generating, by a regressor implemented by the hardware processor, a predicted lung function metric of the user based on the cough feature vector;

wherein the cough detector includes a cough segmenter configured to extract edges of a burst phase of the cough event and reject non-burst portions of the cough event, and the cough feature vector is generated only based upon the burst phase of the cough event.