Cough detecting methods and devices for detecting coughs

• US 10,448,920 B2
• Filed: 09/14/2012
• Issued: 10/22/2019
• Est. Priority Date: 09/15/2011
• Status: Active Grant

• Current Assignee: University of Washington
• Original Assignee: University of Washington
• Inventors: Patel, Shwetak N., Larson, Eric Cooper, Lee, Tien-Jui, Liu, Shih-Yen
• Primary Examiner(s): Cerioni, Daniel L
• Assistant Examiner(s): Warsi, Yasmeen S
• Application Number:
  US14/343,374
• Publication Number:
  US 20140336537A1
• Time in Patent Office: 2,594 Days
• CPC Class Codes: A61B 5/0816 Measuring devices for examining respiratory frequency measuring frequency of electric signals G01R23/00
• CPC Class Codes: A61B 5/0823 Detecting or evaluating cough events
• CPC Class Codes: A61B 7/003 Detecting lung or respiration noise
• CPC Class Codes: A61B 7/04 Electric stethoscopes microphones, acoustic transducers therefor H04R
• CPC Class Codes: G10L 25/66 for extracting parameters related to health condition detecting or measuring for diagnostic purposes A61B5/00

Abstract

Examples of the present invention utilize principal component analysis (PCA) to detect cough sounds in an audio stream. Comparison of all or portions of the audio stream with a cough model may be conducted. The cough model may include a number of basis vectors may be based on initial portions of known coughs. The initial portions may be non-user specific, and accordingly the cough model may be used to detect coughs across individuals. Moreover, examples of the present invention may reconstruct the cough sounds from stored features such that the cough sounds are reconstructed but the reconstruction techniques used may be insufficient to reconstruct speech sounds that may also have been recorded, which may increase user privacy.

Claims

1. A tangible computer readable medium encoded with instructions that, when executed, cause a computing system to;

receive an audio signal from a microphone;

convert at least a portion of the audio into a frequency-based matrix representation;

transform the frequency-based matrix into a lesser dimensional matrix using projections from a set of basis vectors in a cough model to prevent reconstruction of speech sounds from the audio signal;

classify at least a portion of the audio signal as corresponding, to a cough based, at least in part, on the lesser dimensional matrix;

classify multiple portions of the audio signal as respective multiple coughs and the multiple coughs as a cough epoch, based at least in part, on the lesser dimensional matrix;

reconstruct the at least a portion of the audio signal using multiplication of the basis vectors with the lesser dimensional matrix to provide a reconstructed audio signal, wherein the reconstructed audio signal excludes the speech sounds from the audio signal, wherein reconstruction includes implementation of an iterative reconstruction approach including minimizing differences between the frequency based representation of the at least a portion of the audio signal and a reconstructed frequency-based representation;

provide data associated with the multiple portions of the audio signal, wherein the data includes a number of coughs and a number of cough epochs; and

replay the reconstructed audio signal.

2. The computer readable medium of claim 1, wherein the frequency-based matrix representation of the at least a portion of the audio signal is an audio spectrogram.

3. The computer readable medium of claim 1, wherein the basis vectors of the cough model comprise eigenvectors of a covariance matrix of frequency-based matrix representations of audio signals corresponding to known cough sounds.

4. The computer readable medium of claim 1, wherein the basis vectors of the cough model comprise magnitude spectrograms of audio signals corresponding to known cough sounds.

5. The computer readable medium of claim 1, wherein the basis vectors of the cough model comprise complex spectrograms of audio signals corresponding to known cough sounds.

6. The computer readable medium of claim 1, wherein the instructions which, when executed, cause the computing system to classify at least a portion of the audio signal as corresponding to a cough include instructions for calculating a score for one or more of the basis vectors in the cough model.

7. The computer readable medium of claim 6, wherein the instructions that, when executed, cause the computing system to classify at least a portion of the audio signal as corresponding to a cough include instructions for comparing the calculated scores for the one or more basis vectors to threshold values and discarding portions of the transformed audio signal having calculated scores below a predetermined threshold value.

8. The computer readable medium of claim 1, wherein the instructions that, when executed, cause the computing system to classify at least a portion of the audio signal as corresponding to a cough include instructions for implementing a random forest classifier based on a comparison of the lesser dimensional matrix representation of the at least a portion of the audio signal with the basis vector representation of the cough model.

9. The computer readable medium of claim 1, wherein the cough model is a non-user specific cough model.

10. The computer readable medium of claim 1, wherein the computing system comprises a mobile phone.

11. The computer readable medium of claim 1, wherein the cough model is stored in a memory accessible to the computing system.

12. The computer readable medium of claim 1, wherein the instructions, when executed, further cause the computing system to store a duration of the cough, frequency of multiple coughs in the audio signal, number of coughs in the audio signal, cough epochs in the audio signal, or combinations thereof.

20. The computer readable medium of claim 1, wherein the data further includes cough frequency, cough length, and cough quality.

13. A device comprising:

a microphone;

at least one processor; and

a memory encoded with computer readable instructions that, when executed, cause the at least one processor to;

receive an audio signal from a microphone;

convert at least a portion of the audio into a frequency-based matrix representation;

transform the frequency-based matrix into a lesser dimensional matrix using projections from a set of basis vectors in a cough model to prevent reconstruction of speech sounds from the audio signal;

classify a portion of the audio signal as corresponding to a cough based, at least in part, on the lesser dimensional matrix;

classify multiple portions of the audio signal as respective multiple coughs and the multiple coughs as a cough epoch, based at least in part, on the lesser dimensional matrix;

reconstruct the at least a portion of the audio signal using multiplication of the basis vectors with the lesser dimensional matrix to provide a reconstructed audio signal, wherein the reconstructed audio signal excludes the speech sounds from the audio signal, wherein reconstruction includes implementation of an iterative reconstruction approach including minimizing differences between the frequency based representation of the at least a portion of the audio signal and a reconstructed frequency-based representation;

provide data associated with the multiple portions of the audio signal, wherein the data includes a number of coughs and a number of cough epochs; and

replay the reconstructed audio signal.

14. The device of claim 13, wherein the device is a mobile phone.

15. The device of claim 13, wherein the memory is further configured to store the cough model.

16. The device of claim 13, wherein the device further comprises a network interface and wherein the instructions, when executed, further cause the at least one processor to transmit the lesser dimensional matrix to another device for reconstruction.

17. The device of claim 16, wherein the another device is configured to utilize the lesser dimensional matrix to reconstruct the audio corresponding to the cough.

18. A method for reconstructing cough sounds, the method comprising:

A microphone;

at least one processor;

a memory encoded with computer readable instructions that, when executed, cause the at least one processor to;

receive an audio signal from a microphone;

convert at least a portion of the audio into a frequency-based matrix representation;

transform the frequency-based matrix into a lesser dimensional matrix using projections from a set of basis vectors in a cough model to prevent reconstruction of speech sounds from the audio signal;

classify a portion of the audio signal as corresponding to a cough based, at least in part, on the lesser dimensional matrix;

classify multiple portions of the audio signal as respective multiple coughs and the multiple coughs as a cough epoch, based at least in part, on the lesser dimensional matrix;

reconstruct the at least a portion of the audio signal using multiplication of the basis vectors with the lesser dimensional matrix to provide a reconstructed audio signal, wherein the reconstructed audio signal excludes the speech sounds from the audio signal, wherein reconstruction includes implementation of an iterative reconstruction approach including minimizing differences between the frequency based representation of the at least a portion of the audio signal and a reconstructed frequency-based representation;

provide data associated with the multiple portions of the audio signal, wherein the data includes a number of coughs and a number of cough epochs; and

replay the reconstructed audio signal.

19. The method of claim 18, wherein the basis vectors are based on selected portions of a cough.