Phonologically-based biomarkers for major depressive disorder

Patent metadata

Field	Value
Publication number	US-9936914-B2
Application number	US-201715670064-A
Country	US
Kind code	B2
Filing date	Aug 7, 2017
Priority date	Aug 2, 2011
Publication date	Apr 10, 2018
Grant date	Apr 10, 2018

Abstract

A system and a method for assessing a condition in a subject. Phones from speech of the subject are recognized, one or more prosodic or speech-excitation-source features of the phones are extracted, and an assessment of a condition of the subject, is generated based on a correlation between the features of the phones and the condition.

First claim

What is claimed is:

1. A computer-implemented method for presenting a measurement of a physical or psychological disorder of a subject determined from the subject's production of a speech signal, the method comprising:
receiving audio data representing an audio signal produced by a microphone in response to the speech signal produced by the subject;
using a computer-implement speech recognizer to segment the audio data into a plurality of segments of the audio data, each segment of the audio data representing a corresponding time interval of the speech signal, wherein each segment of the audio data is associated is a corresponding speech unit of a predefined plurality of speech units, at least one speech unit corresponds to multiple segments of the plurality of segments, and a represented plurality of speech units comprises speech units of the plurality of speech units that correspond to at least one of the segments of the audio data;
processing the segments of the audio data to produce respective values of segment features, the segment features for a segment characterizing the subject's production of the speech unit correspond to the segment;
for each represented speech unit of the represented plurality of speech units, combining the values of the segment features for segments of the audio data corresponding to the represented speech unit to determine values of speech-unit features corresponding to the represented speech unit;
forming a feature representation of the audio data from the values of the speech-unit features corresponding to each of the represented speech units;
processing the feature representation of the audio data according to values of a plurality of numerical configuration parameters to provide one or more disorder indicators, wherein the numerical configuration parameters are formed from audio data for a plurality speech signals, each speech signal produced by a corresponding subject and data indicating presence of one or more disorders of the subject corresponding to each of the speech signals, each of the disorder indicators corresponds to a physical or psychological disorder; and
determining output data from the one or more numerical disorder indicators and outputting the data to a user to indicate presence of one or more disorders of the plurality of disorders in the subject.

2. The method of claim 1 wherein the speech units comprise linguistically-based speech units.

3. The method of claim 2 wherein the linguistically-based speech units comprise a unit defined by a particular phone type.

4. The method of claim 1 wherein using the computer-implement speech recognizer to segment the audio data comprises using a Hidden Markov Model speech recognizer.

5. The method of claim 1 wherein the segment features comprise segment duration and the speech-unit features comprise average segment duration, wherein combining the values of the segment features for segments of the audio data corresponding to the represented speech unit comprises averaging the durations for said segments to determine the average segment duration.

6. The method of claim 1 wherein the segment features comprise prosodic or speech-excitation-source features.

7. The method of claim 6 wherein prosodic or speech-excitation-source features comprise at least one of fundamental frequency, pitch, amplitude, and RMS power.

8. The method of claim 1 wherein the segment features comprise an energy spread around a centroid within the segment.

9. The method of claim 1 further comprising determining the numerical configuration parameters, including:
receiving audio data representing a plurality of speech signals produced by the respective subjects;
receiving data indicating presence of one or more disorders of the plurality of disorders output for each of the subjects;
processing the audio data to determine values speech-unit features and a feature representation corresponding to each of the speech signals; and
determining the configuration parameters to configure a disorder predictor for predicting the one or more disorders from a feature representation.

10. The method of claim 9 wherein processing the feature representation according to a plurality of numerical configuration parameters comprises providing the feature representation to the disorder predictor to produce the one or more disorder indicators.

11. The method of claim 9 wherein determining the numerical configuration parameters comprises computing correlations between the speech-unit features and the disorder indicators.

12. The method of claim 1 wherein the method is applied for remote assessment of a subject.

Assignees

• Massachusetts Inst Technology

Inventors

• Quatieri Jr Thomas F
• Malyska Nicolas
• Trevino Andrea Carolina

Classifications

• A61B5/4803 Primary

  Speech analysis specially adapted for diagnostic purposes · CPC title

• G10L25/18

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

• G10L25/06

  the extracted parameters being correlation coefficients · CPC title

• A61B5/7267

  involving training the classification device · CPC title

• G10L25/21

  the extracted parameters being power information · CPC title