US 11,672,472 B2
Methods and systems for estimation of obstructive sleep apnea severity in wake subjects by multiple speech analyses
Yaniv Zigel, Omer (IL); Dvir Ben Or, Gedera (IL); Ariel Tarasiuk, Meitar (IL); and Eliran Dafna, Beer-Sheva (IL)
Assigned to B.G. NEGEV TECHNOLOGIES AND APPLICATIONS LTD., AT BEN-GURION UNIVERSITY, Beer-Sheva (IL); and MOR RESEARCH APPLICATIONS LTD., Tel Aviv (IL)
Appl. No. 16/316,395
Filed by B.G. NEGEV TECHNOLOGIES AND APPLICATIONS LTD., AT BEN-GURION UNIVERSITY, Beer-Sheva (IL); and MOR RESEARCH APPLICATIONS LTD., Tel Aviv (IL)
PCT Filed Jul. 10, 2017, PCT No. PCT/IL2017/050779
§ 371(c)(1), (2) Date Jan. 9, 2019,
PCT Pub. No. WO2018/011794, PCT Pub. Date Jan. 18, 2018.
Claims priority of provisional application 62/360,400, filed on Jul. 10, 2016.
Prior Publication US 2019/0298271 A1, Oct. 3, 2019
Int. Cl. A61B 5/00 (2006.01); G16H 50/30 (2018.01); A61B 7/00 (2006.01); G16H 50/20 (2018.01); G16H 50/50 (2018.01)
CPC A61B 5/4818 (2013.01) [A61B 5/4803 (2013.01); A61B 5/4842 (2013.01); A61B 5/7267 (2013.01); A61B 5/7275 (2013.01); A61B 5/7278 (2013.01); A61B 7/003 (2013.01); G16H 50/30 (2018.01); G16H 50/20 (2018.01); G16H 50/50 (2018.01)] 16 Claims
OG exemplary drawing
 
1. A method for determining apnea-hypopnea index (AHI) estimation of a wake subject, said method comprising:
obtaining an audio recorded phonogram signal comprising a speech segment and an audio recorded phonogram signal comprising a sustained-vowel segment;
framing said signal comprising a sustained-vowel segment into frames;
framing said signal comprising a speech segment into a set of long term frames and into a set of short term frames;
generating one or more feature parameters for each short term frame, that are associated with short term characteristics and that are calculated according to the short term frame signal or according to a signal generated from the short term frame signal;
generating one or more feature parameters for each long term frame, that are associated with long term characteristics and that are calculated according to the long term frame signal or according to a signal generated from the long term frame signal;
generating one or more feature parameters for each sustained-vowel frame, that are associated with sustained-vowel characteristics and that are calculated according to the sustained-vowel frame signal or according to a signal generated from the sustained-vowel frame signal;
applying a first computing system preformed model on the generated short term feature parameters to obtain a short term score;
applying a second computing system preformed model on the generated long term feature parameters to obtain a long term score;
applying a third computing system preformed model on the generated sustained-vowel feature parameters to obtain a sustained-vowel score; and
applying a fusing model on the short term score and on the long term score and on the sustained-vowel score to obtain said subject's apnea-hypopnea index estimation;
wherein the sustained-vowel segment comprises a subject's uttering of a vowel sound for a period of between 1 and 10 seconds; and
wherein:
(a) said one or more sustained vowel feature parameters comprise in a male subject at least one of: a “u:” phoneme arithmetic mean of soft phonation index (SPI), an “e” phoneme geometric mean of spectral flatness, a “u:” phoneme zero crossing rate of a first time derivative of Thomson multitaper mel-frequency 1st cepstral coefficient ΔMFCC1, an “custom character” phoneme zero crossing rate of a first time derivative of Thomson multitaper mel-frequency 2nd cepstral coefficient ΔMFCC2, or an “custom character” phoneme inter-quartile range of a first time derivative of Thomson multitaper mel-frequency 4th cepstral coefficient ΔMFCC4; or
(b) said one or more sustained vowel feature parameters in a female subject comprise at least one of an “custom character” phoneme skewness of Thomson multitaper mel-frequency 5th cepstral coefficient MFCC5, an “a:” phoneme maximum of spectral flatness, an “e” phoneme arithmetic mean of 3-point amplitude perturbation quotient APQ3, an “a:” phoneme quadratic mean of shimmer, or an “i” phoneme second quartile of Thomson multitaper mel-frequency 12th cepstral coefficient MFCC12.