Model learning device, estimating device, methods therefor, and program
State-of-satisfaction change pattern models each including a set of transition weights in state sequences of the states of satisfaction are obtained for predetermined change patterns of the states of satisfaction, and a state-of-satisfaction estimation model for obtaining the posteriori probability of the utterance feature amount given the state of satisfaction of an utterer is obtained by using the utterance-for-learning feature amount and a correct value of the state of satisfaction of an utterer who gave an utterance for learning corresponding to the utterance-for-learning feature amount. By using the input utterance feature amount and the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model, an estimated value of the state of satisfaction of an utterer who gave an utterance corresponding to the input utterance feature amount is obtained.
1. A model learning device comprising processing circuitry configured to:
obtain, for a plurality types of predetermined change patterns of a plurality of states of satisfaction, a plurality of state-of-satisfaction change pattern models each including a set of transition weights in a plurality of state sequences of the plurality of states of satisfaction of each of the predetermined change patterns by using state-of-satisfaction change pattern correct values indicating correct values of change patterns of state of satisfactions of an utterer in a conversation and state-of-satisfaction correct values, each indicating a correct value of the state of satisfaction of the utterer at the time of each utterance in the conversation, and output the state-of-satisfaction change pattern models;
obtain, by using an utterance-for-learning feature amount and a correct value of a state of satisfaction of an utterer who gave an utterance for learning corresponding to the utterance-for-learning feature amount, a state-of-satisfaction estimation model for obtaining a posteriori probability of an utterance feature amount given a state of satisfaction of an utterer, and output the state-of-satisfaction estimation model; wherein
the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model are input to an estimating device,
the estimating device
receives a plurality of utterances given by a particular utterer in a conversation,
detects one or more voice activities in each of the plurality of utterances received, and
by using an input utterance feature amount extracted, based on the detected one or more voice activities, and the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model, obtains an estimated value of a state of satisfaction of the particular utterer who provided the plurality of utterances to the estimating device and outputs the estimated value.
2. The model learning device according to claim 1 , wherein
the states of satisfaction include any one of states: satisfaction, average, and dissatisfaction, and
the change patterns include any one of
(1) a pattern in which the state of satisfaction changes from average to satisfaction,
(2) a pattern in which the state of satisfaction changes from average to dissatisfaction and then changes to satisfaction,
(3) a pattern in which the state of satisfaction changes from dissatisfaction to satisfaction,
(4) a pattern in which average continues,
(5) a pattern in which the state of satisfaction changes from average to dissatisfaction and then changes to average,
(6) a pattern in which dissatisfaction continues,
(7) a pattern in which the state of satisfaction changes from average to dissatisfaction,
(8) a pattern in which the state of satisfaction changes from dissatisfaction to average, and
(9) a pattern in which satisfaction continues.
3. The model learning device according to claim 1 , wherein
a state-of-satisfaction change pattern model structure is the state sequence of the states of satisfaction, and
the processing circuitry obtains, for the plurality types of the change patterns, the plurality of the state-of-satisfaction change pattern models by using the same state-of-satisfaction change pattern model structure for all the change patterns and outputs the state-of-satisfaction change pattern models.
4. An estimating device comprising processing circuitry comprising processing circuitry configured to:
receive the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model of any one of claims 1 to 3 ;
receive a plurality of utterances given by a particular utterer in a conversation;
detect one or more voice activities in each of the plurality of utterances; and
by using an input utterance feature amount, based on the detected one or more voice activities, and the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model, obtain an estimated value of a state of satisfaction of an utterer who provided the plurality of utterances to the estimating device and output the estimated value.
5. A model learning method of a model learning device, the model learning method, executed by processing circuitry, comprising:
obtaining, for a plurality types of predetermined change patterns of a plurality of states of satisfaction, a plurality of state-of-satisfaction change pattern models each including a set of transition weights in a plurality of state sequences of the plurality of states of satisfaction of each of the predetermined change patterns by using state-of-satisfaction change pattern correct values indicating correct values of change patterns of state of satisfactions of an utterer in a conversation and state-of-satisfaction correct values, each indicating a correct value of the state of satisfaction of the utterer at the time of each utterance in the conversation, and outputting the state-of-satisfaction change pattern models; and
obtaining, by using an utterance-for-learning feature amount and a correct value of a state of satisfaction of an utterer who gave an utterance for learning corresponding to the utterance-for-learning feature amount, a state-of-satisfaction estimation model for obtaining a posteriori probability of an utterance feature amount given a state of satisfaction of an utterer, and outputting the state-of-satisfaction estimation model, wherein
the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model are input to an estimating device, the estimating device receives a plurality of utterances given by a particular utterer in a conversation,
detects one or more voice activities in each of the plurality of utterances, and
by using an input utterance feature amount extracted, based on the detected one or more voice activities, and the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model, obtains an estimated value of a state of satisfaction of the particular utterer who provided the plurality of utterances to the estimating device and outputs the estimated value.
6. The model learning method according to claim 5 , wherein
a state-of-satisfaction change pattern model structure is the state sequence of the states of satisfaction, and
the state-of-satisfaction change pattern model learning step obtains, for the plurality types of the change patterns, the plurality of the state-of-satisfaction change pattern models by using the same state-of-satisfaction change pattern model structure for all the change patterns and outputs the state-of-satisfaction change pattern models.
7. An estimating method of an estimating device, the estimating method, executed by processing circuitry, comprising:
receiving the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model of claim 5 or 6 ;
receiving a plurality of utterances given by a particular utterer in a conversation;
detecting one or more voice activities in each of the plurality of utterances; and
by using an input utterance feature amount, based on the detected one or more voice activities, and the state-of-satisfaction change pattern models and the state-of-satisfaction estimation model, obtaining an estimated value of a state of satisfaction of an utterer who provided the plurality of utterances to the estimating device and outputting the estimated value.
8. A non-transitory computer-readable recording medium storing a program for causing a computer to execute the model learning device according to claim 5 or 6 .
9. A non-transitory computer-readable recording medium storing a program for causing a computer to execute the estimating method according to claim 7 .