System and method for assessing suicide risk of a patient based upon non-verbal characteristics of voice data

Patent metadata

Field	Value
Publication number	US-9691411-B2
Application number	US-201414893253-A
Country	US
Kind code	B2
Filing date	May 23, 2014
Priority date	May 24, 2013
Publication date	Jun 27, 2017
Grant date	Jun 27, 2017

Abstract

A method for assessing suicide risk for a human subject including receiving recorded voice data of the subject; and classifying the subject as suicidal or non-suicidal based upon a computerized analysis of one or more nonverbal characteristics of the speech data, especially features associated with a breathy phonation type. The analysis of the nonverbal characteristics of the voice data can include an analysis of acoustic characteristics of speech, and/or an analysis of prosodic and voice quality-related features of the voice data. Related apparatus, systems, techniques and articles are also described.

First claim

The invention claimed is:

1. A computer-implemented method for assessing suicide risk in a human subject, the method comprising the steps of:
receiving voice data of the subject; and
classifying the subject as suicidal or non-suicidal based upon a computerized analysis of one or more nonverbal characteristics of the voice data performed on a support vector machine (SVM) that utilizing hidden Markov models (HMMs),
the nonverbal characteristics of the voice data including prosodic and voice quality-related features of the voice data selected from one or more of an acoustic measure of energy in dB (en, en slope ), an acoustic measure of fundamental frequency (f 0 ), an acoustic measure of peak slope (peak), and an acoustic measure of spectral stationarity (ss),
the prosodic and voice quality features including an acoustic measure of Liljencrants-Fant (LF) model parameters from time domain estimation methods, including three time-based parameters (R a , R k and R g ) an amplitude parameter (EE), an Open Quotient (OQ), and a parameter characterizing the basic shape of the LF model (R d ) and
the classifying including classifying the subject as being suicidal based upon the voice data exhibiting the characteristics of a breathy phonation type.

2. The method of claim 1 , wherein the characteristics of a breathy phonation type are determined based upon one or more acoustic features selected from the group consisting of: Open Quotient (OQ), Normalized Amplitude Quotient (NAQ) and peak slope (peak).

3. The method of claim 2 , wherein the characteristics of a breathy phonation type are determined based upon one or more acoustic features selected from relatively larger R k values and relatively smaller R g values.

4. A system for assessing suicide risk for a human subject, the system comprising at least one processor, and a non-transitory computer-readable medium containing instructions stored thereon which when executed by the at least one processor, perform a method of assessing suicide risk for the human subject, the method comprising:
receiving voice data of the subject; and
classifying the subject as suicidal or non-suicidal based upon a computerized analysis of one or more nonverbal characteristics of the voice data performed on a support vector machine (SVM) utilizing hidden Markov models (HMMs),
the nonverbal characteristics of the voice data including prosodic and voice quality-related features of the voice data selected from one or more of an acoustic measure of energy in dB (en, en slope ), an acoustic measure of fundamental frequency (f 0 ), an acoustic measure of peak slope (peak), and an acoustic measure of spectral stationarity (ss),
the prosodic and voice quality features including an acoustic measure of Liljencrants-Fant (LF) model parameters from time domain estimation methods, including three time-based parameters (R a , R k , R g ) an amplitude parameter (EE), an Open Quotient (OQ), and a parameter characterizing the basic shape of the LF model (R d ) and
the classifying including classifying the subject as being suicidal based upon the voice data exhibiting the characteristics of a breathy phonation type.

5. The system of claim 4 , wherein the characteristics of a breathy phonation type are determined based upon one or more acoustic features selected from the group consisting of: Open Quotient (OQ), Normalized Amplitude Quotient (NAQ) and peak slope (peak) features.

6. The system of claim 5 , wherein the characteristics of a breathy phonation type are determined based upon one or more acoustic features selected from relatively larger R k values, and relatively smaller R g values.

Assignees

• Children'S Hospital Medical Center
• Univ Southern California

Inventors

• Scherer Stefan
• Pestian John P
• Morency Louis-Philippe

Classifications

• G10L25/63 Primary

  for estimating an emotional state · CPC title

• G10L25/21

  the extracted parameters being power information · CPC title

• G10L17/26

  Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices · CPC title

• G10L17/16

  Hidden Markov models [HMM] · CPC title

• G10L25/18

  the extracted parameters being spectral information of each sub-band · CPC title