Diagnostic techniques based on speech models
A method includes obtaining one or more speech models, each model including one or more acoustic states and, provided that the model includes multiple acoustic states, allowed transitions therebetween. The method further includes receiving a speech sample produced by a subject while a physiological state of the subject was unknown. The method further includes mapping at least one sample portion of the speech sample to a respective one of the speech models, by computing a plurality of feature vectors quantifying acoustic features of different respective portions of the sample portion, and mapping the feature vectors to respective acoustic states included in the speech model such that a total distance between the feature vectors and the respective acoustic states is minimized. The method further includes, in response to mapping the sample portion to the speech model, communicating an output indicating the physiological state of the subject. Other embodiments are also described.
1. A system, comprising:
circuitry; and
one or more processors, configured to cooperatively carry out a process that includes:
obtaining one or more speech models, each of the speech models including (i) one or more acoustic states associated with respective local distance functions such that, given any acoustic feature vector within a domain of the local distance functions, the local distance function of each acoustic state returns a local distance indicating a degree of correspondence between the given acoustic feature vector and the acoustic state, and (ii) provided that the speech model includes multiple acoustic states, allowed transitions between the acoustic states,
receiving, via the circuitry, a speech sample produced by a subject while a physiological state of the subject was unknown,
mapping at least one sample portion of the speech sample to a respective one of the speech models, by:
computing a plurality of feature vectors that quantify acoustic features of different respective portions of the sample portion, and
based on the local distance functions and on the allowed transitions included in the speech model, mapping the feature vectors to respective ones of the acoustic states included in the speech model such that a total distance between the feature vectors and the respective ones of the acoustic states is minimized, the total distance being based on respective local distances between the feature vectors and the respective ones of the acoustic states, and
in response to mapping the sample portion to the speech model, communicating an output indicating the physiological state of the subject.
2. The system according to claim 1 , wherein the speech models were constructed from reference speech produced in a known state with respect to a physiological condition.
3. The system according to claim 2 , wherein at least some of the reference speech was produced by the subject.
4. The system according to claim 2 , wherein the reference speech includes reference free speech.
5. The system according to claim 1 ,
wherein the one or more speech models include multiple speech models representing different respective speech units,
wherein the speech sample includes free speech,
wherein the sample portion includes one of the speech units, and
wherein mapping the sample portion to the respective one of the speech models includes mapping the sample portion to the speech model representing the speech unit included in the sample portion.
6. The system according to claim 5 , wherein obtaining the speech models includes obtaining the speech models by:
identifying the speech units in reference speech, and
based on the speech units, constructing the speech models.
7. The system according to claim 1 , wherein the total distance is based on a sum of the respective local distances.
8. A method, comprising:
obtaining, by a processor, one or more speech models, each of the speech models including (i) one or more acoustic states associated with respective local distance functions such that, given any acoustic feature vector within a domain of the local distance functions, the local distance function of each acoustic state returns a local distance indicating a degree of correspondence between the given acoustic feature vector and the acoustic state, and (ii) provided that the speech model includes multiple acoustic states, allowed transitions between the acoustic states;
receiving, by the processor, a speech sample produced by a subject while a physiological state of the subject was unknown;
mapping, by the processor, at least one sample portion of the speech sample to a respective one of the speech models, by:
computing a plurality of feature vectors that quantify acoustic features of different respective portions of the sample portion, and
based on the local distance functions and on the allowed transitions included in the speech model, mapping the feature vectors to respective ones of the acoustic states included in the speech model such that a total distance between the feature vectors and the respective ones of the acoustic states is minimized, the total distance being based on respective local distances between the feature vectors and the respective ones of the acoustic states; and
in response to mapping the sample portion to the speech model, communicating, by the processor, an output indicating the physiological state of the subject.
9. The method according to claim 8 , wherein the speech models were constructed from reference speech produced in a known state with respect to a physiological condition.
10. The method according to claim 9 , wherein at least some of the reference speech was produced by the subject.
11. The method according to claim 9 , wherein the reference speech includes reference free speech.
12. The method according to claim 8 ,
wherein the one or more speech models include multiple speech models representing different respective speech units,
wherein the speech sample includes free speech,
wherein the sample portion includes one of the speech units, and
wherein mapping the sample portion to the respective one of the speech models comprises mapping the sample portion to the speech model representing the speech unit included in the sample portion.
13. The method according to claim 12 , wherein obtaining the speech models comprises obtaining the speech models by:
identifying the speech units in reference speech, and
based on the speech units, constructing the speech models.
14. The method according to claim 8 , wherein the total distance is based on a sum of the respective local distances.
15. A computer software product comprising a tangible non-transitory computer-readable medium in which program instructions are stored, which instructions, when read by a processor, cause the processor to:
obtain one or more speech models, each of the speech models including (i) one or more acoustic states associated with respective local distance functions such that, given any acoustic feature vector within a domain of the local distance functions, the local distance function of each acoustic state returns a local distance indicating a degree of correspondence between the given acoustic feature vector and the acoustic state, and (ii) provided that the speech model includes multiple acoustic states, allowed transitions between the acoustic states,
receive a speech sample produced by a subject while a physiological state of the subject was unknown,
map at least one sample portion of the speech sample to a respective one of the speech models, by:
computing a plurality of feature vectors that quantify acoustic features of different respective portions of the sample portion, and
based on the local distance functions and on the allowed transitions included in the speech model, mapping the feature vectors to respective ones of the acoustic states included in the speech model such that a total distance between the feature vectors and the respective ones of the acoustic states is minimized, the total distance being based on respective local distances between the feature vectors and the respective ones of the acoustic states, and
in response to mapping the sample portion to the speech model, communicate an output indicating the physiological state of the subject.
16. The computer software product according to claim 15 , wherein the speech models were constructed from reference speech produced in a known state with respect to a physiological condition.
17. The computer software product according to claim 16 , wherein at least some of the reference speech was produced by the subject.
18. The computer software product according to claim 16 , wherein the reference speech includes reference free speech.
19. The computer software product according to claim 15 ,
wherein the one or more speech models include multiple speech models representing different respective speech units,
wherein the speech sample includes free speech,
wherein the sample portion includes one of the speech units, and
wherein the instructions cause the processor to map the sample portion to the speech model representing the speech unit included in the sample portion.
20. The computer software product according to claim 19 , wherein the instructions cause the processor to obtain the speech models by:
identifying the speech units in reference speech, and based on the speech units, constructing the speech models.
21. The computer software product according to claim 15 , wherein the total distance is based on a sum of the respective local distances.