Converting a sequence of speech records of a human subject into a sequence of indicators of a physiological state of the subject
A system includes a memory and a processor. The memory is configured to store a machine learning (ML) model trained using a plurality of sequences of speech records of humans having each at least one of a respective known physiological state, psychological state and emotional state. The processor is configured to (i) receive a sequence of speech records of a human subject, (ii) apply the trained ML model to infer from the sequence of speech records of the human subject a sequence of one or more indicators indicative of at least one: a physiological state, a psychological state, and an emotional state of the human subject, and (iii) make the indicators available.
1 . A system for evaluating at least one state selected from the group consisting of: a physiological state, a psychological state, and an emotional state, the system comprising:
a memory configured to store a machine learning (ML) model trained using a plurality of training sequences of speech records, each of the speech records having been produced by a respective human while the state of the human was known; and
a processor, which is configured to:
receive a time sequence of speech records of a human subject;
apply the trained ML model to the time sequence of speech records so as to convert the time sequence of speech records, via a sequence-to-sequence conversion, to a sequence of indicators indicative of the state of the human subject at different respective time points in a time period; and
output the sequence of indicators.
2 . The system according to claim 1 , wherein the time sequence of speech records is a sequence of speech utterances associated with respective recording times, and wherein the recording times are also inputs to the trained ML model.
3 . The system according to claim 1 , wherein the processor is configured to apply the ML model by representing the time sequence of speech records by a sequence of embedding vectors and applying the model to the embedding vectors.
4 . The system according to claim 3 , wherein each embedding vector represents a respective subset of the speech records, independently of other embedding vectors.
5 . The system according to claim 3 , wherein the processor is further configured to aggregate a portion of the embedding vectors into a single aggregated embedding vector and to apply the model to a set of the embedding vectors comprising the aggregated embedding vector.
6 . The system according to claim 1 , wherein at least some of the indicators comprise at least one element selected from the group consisting of; a label selected from a set of labels associated with a predefined set of states, and an estimated level of confidence in the indication.
7 . The system according to claim 1 , wherein the processor is configured to output the sequence of indicators by at least one action selected from the group consisting of: sending the indicators over a network to a user's device, providing an interface that allows the user to access them, sending an alert message when the output sequence of indicators fits a certain pattern, and presenting the indicators graphically.
8 . The system according to claim 1 , wherein the known state is with respect to a medical condition and wherein the indicators are clinical indicators.
9 . The system according to claim 8 , wherein the medical condition is selected from a list of medical conditions consisting of: cardio-vascular disease, pulmonary disease, respiratory disease, neurological disease, psychological disease, congestive heart failure, coronary heart disease, heart arrhythmia, chronic obstructive pulmonary disease (COPD), interstitial lung disease, asthma, Hypocarbia, Hypercarbia, altitude sickness, Parkinson disease, Alzheimer disease, bipolar disorder mild cognitive impairment and depression.
10 . The system according to claim 8 , wherein the clinical indicators are indicative of a presence of excess fluid in the chest.
11 . The system according to claim 1 , wherein the indicators provide an estimate of time left until the state reaches a certain stage.
12 . The system according to claim 1 , where one or more of the indicators comprise estimates of measurable parameters of the human subject that are indicative of the state of the human subject.
13 . The system according to claim 1 , wherein one or more of the indicators indicate whether the human subject is stable or unstable with respect to the state.
14 . The system according to claim 1 , wherein the trained ML model comprises a trained neural network (NN) model.
15 . The system according to claim 14 , wherein the trained NN model comprises a Transformer model comprising an Attention model.
16 . The system according to claim 14 , wherein the trained NN model comprises a recursive neural network (RNN) model that adjusts the weights of the RNN during inference.
17 . The system according to claim 1 , wherein the processor is configured to apply the trained ML model to the time sequence of the speech records by grouping the sequence of the speech records into a sequence of sessions of the speech records and applying the ML model to the sequence of sessions of the speech records.
18 . A system for evaluating at least one state selected from the group consisting of: a physiological state, a psychological state, and an emotional state, the system comprising:
a memory configured to store a machine learning (ML) model; and
a processor, which is configured to:
receive a plurality of training sequences of speech records, each of the speech records having been produced by a respective human while the state of the human was known; and
using the plurality of training sequences of speech records, train the ML model to convert a time sequence of speech records of a human subject, via a sequence-to-sequence conversion, to a sequence of indicators indicative of the state of the human subject at different respective time points in a time period.
19 . A method for evaluating at least one state selected from the group consisting of: a physiological state, a psychological state, and an emotional state, the method comprising:
storing a machine learning (ML) model trained using a plurality of training sequences of speech records, each of the speech records having been produced by a respective human while the state of the human was known;
receiving a time sequence of speech records of a human subject;
applying the trained ML model to the time sequence of speech records so as to convert the time sequence of speech records, via a sequence-to-sequence conversion, to a sequence of indicators indicative of the state of the human subject at different respective time points in a time period; and
outputting the sequence of indicators.
20 . The method according to claim 19 , wherein the time sequence of speech records is a sequence of speech utterances associated with respective recording times, and wherein the recording times are also inputs to the trained ML model.
21 . The method according to claim 19 , wherein applying the ML model comprises representing the time sequence of speech records by a sequence of embedding vectors and applying the model to the embedding vectors.
22 . The method according to claim 21 , wherein each embedding vector represents a respective subset of the speech records, independently of other embedding vectors.
23 . The method according to claim 21 , and comprising aggregating a portion of the embedding vectors into a single aggregated embedding vector and applying the model to a set of the embedding vectors comprising the aggregated embedding vector.
24 . The method according to claim 19 , wherein at least some of the indicators comprise at least one element selected from the group consisting of; a label selected from a set of labels associated with a predefined set of states, and an estimated level of confidence in the indication.
25 . The method according to claim 19 , wherein outputting the sequence of indicators comprises at least one action selected from the group consisting of: sending the indicators over a network to a user's device, providing an interface that allows the user to access them, sending an alert message when the output sequence of indicators fits a certain pattern, and presenting the indicators graphically.
26 . The method according to claim 19 , wherein the known state is with respect to a medical condition and wherein the indicators are clinical indicators.
27 . The method according to claim 26 , wherein the medical condition is selected from a list of medical conditions consisting of: cardio-vascular disease, pulmonary disease, respiratory disease, neurological disease, psychological disease, congestive heart failure, coronary heart disease, heart arrhythmia, chronic obstructive pulmonary disease (COPD), interstitial lung disease, asthma, Hypocarbia, Hypercarbia, altitude sickness, Parkinson disease, Alzheimer disease, bipolar disorder mild cognitive impairment and depression.
28 . The method according to claim 26 , wherein the clinical indicators are indicative of a presence of excess fluid in the chest.
29 . The method according to claim 19 , wherein the indicators provide an estimate of time left until the state reaches a certain stage.
30 . The method according to claim 19 , where one or more of the indicators comprise estimates of measurable parameters of the human subject that are indicative of the state of the human subject.
31 . The method according to claim 19 , wherein one or more of the indicators indicate whether the human subject is stable or unstable with respect to the state.
32 . The method according to claim 19 , wherein the trained ML model comprises a trained neural network (NN) model.
33 . The method according to claim 32 , wherein the trained NN model comprises a Transformer model comprising an Attention model.
34 . The method according to claim 32 , wherein the trained NN model comprises a recursive neural network (RNN) model that adjusts the weights of the RNN during inference.
35 . The method according to claim 19 , and comprising applying the trained ML model to the time sequence of the speech records by grouping the sequence of the speech records into a sequence of sessions of the speech records and applying the ML model to the sequence of sessions of the speech records.
36 . A method for evaluating at least one state selected from the group consisting of: a physiological state, a psychological state, and an emotional state, the method comprising:
storing a machine learning (ML) model;
receiving a plurality of training sequences of speech records, each of the speech records having been produced by a respective human while the state of the human was; and
using the plurality of training sequences of speech records, training the ML model to convert a time sequence of speech records of a human subject, via a sequence-to-sequence conversion, to a sequence of indicators indicative of the state of the human subject at different respective time points in a time period.