CLOUD-BASED DEVICE AND OPERATING METHOD THEREFOR
This disclosure provides a cloud-based device and an operating method thereof. The device includes an analysis apparatus, where the analysis apparatus includes: a first HMM analyzer, configured to respectively receive a scenario input signal, an audience expression input signal, and an audience voice input signal, use them as observable sequences of a first HMM, and deduce a hidden state change sequence of the first HMM; an emotional state HMM analyzer, configured to use the hidden state change sequence of the first HMM as an observable sequence of an emotional state HMM, and deduce a hidden state change sequence of the emotional state HMM; and a voice signal processing unit; and a decision apparatus, configured to select humorous behavior based on the hidden state change sequence of the emotional state HMM, and consolidate a humorous behavior instruction and the standard instruction to obtain a final output instruction.
1 . Cloud-based device, comprising:
an analysis apparatus, wherein the analysis apparatus comprises:
a first HMM analyzer, configured to respectively receive a scenario input signal, an audience expression input signal, and an audience voice input signal, use them as observable sequences of a first HMM, and deduce a hidden state change sequence of the first HMM based on an observable sequence probability maximization criterion, wherein the hidden state change sequence of the first HMM comprises a scenario hidden state change sequence, an audience expression hidden state change sequence, and an audience voice hidden state change sequence;
an emotional state HMM analyzer, configured to receive the scenario hidden state change sequence, the audience expression hidden state change sequence, and the audience voice hidden state change sequence, use them as observable sequences of an emotional state HMM, and deduce a hidden state change sequence of the emotional state HMM based on the observable sequence probability maximization criterion; and
a voice signal processing unit, configured to identify the audience voice input signal, and output a standard instruction based on an identification result; and
a decision apparatus, configured to receive the hidden state change sequence of the emotional state HMM and the standard instruction, select humorous behavior based on the hidden state change sequence of the emotional state HMM, and consolidate a humorous behavior instruction and the standard instruction to obtain a final output instruction.
2 . The cloud-based device according to claim 1 , wherein the first HMM analyzer further comprises a scenario HMM analyzer, an audience expression HMM analyzer, and an audience voice HMM analyzer, and the scenario HMM analyzer, the audience expression HMM analyzer, and the audience voice HMM analyzer are connected in a series or parallel manner.
3 . The cloud-based device according to claim 1 , wherein the decision apparatus comprises:
a humorous behavior selection unit, configured to perform probability analysis on the hidden state change sequence of the emotional state HMM, select the humorous behavior, and send the humorous behavior instruction; and
a consolidation unit, configured to consolidate the humorous behavior instruction and the standard instruction to obtain the final output instruction, wherein
an output end of the emotional state HMM analyzer is connected to an input end of the humorous behavior selection unit, an output end of the humorous behavior selection unit is connected to an input end of the consolidation unit, and an output end of the voice signal processing unit is connected to the input end of the consolidation unit.
4 . The cloud-based device according to claim 3 , wherein the consolidation further comprises:
when the humorous behavior instruction is one of “telling a joke”, “reading interesting news”, “performing a funny action”, and “singing”, the consolidation unit selects an optimal humor output instruction by searching a cloud database and/or accessing the Internet with reference to the audience voice input signal, and uses the optimal humor output instruction and the standard instruction as the final output instruction, wherein the optimal humor output instruction is an instruction most matching an emotional state of a target audience.
5 . The cloud-based device according to claim 4 , wherein a related policy for selecting the humorous behavior and the optimal humor output instruction is correspondingly adjusted by using feedback information obtained based on continuous interaction with the target audience.
6 . The cloud-based device according to claim 3 , wherein the probability analysis comprises: calculating, by the humorous behavior selection unit, humorous behavior set probability distribution by using a preset probability transition matrix from an emotional state to a humorous behavior set.
7 . The cloud-based device according to claim 1 , wherein deducing the hidden state change sequence based on the observable sequence probability maximization criterion is implemented by using a Viterbi algorithm.
8 . The cloud-based device according to claim 1 , wherein an output end of the first HMM analyzer is connected to an input end of the emotional state HMM analyzer.
9 . The cloud-based device according to claim 2 , wherein one or more of output ends of the scenario HMM analyzer, the audience expression HMM analyzer, and the audience voice HMM analyzer are connected to an input end of the emotional state HMM analyzer.
10 . The cloud-based device according to claim 1 , wherein an output end of the analysis apparatus is connected to an input end of the decision apparatus.
11 . The cloud-based device according to claim 1 , wherein the device further comprises a first transceiver, an output end of the first transceiver is connected to an input end of the analysis apparatus, and an output end of the decision apparatus is connected to an input end of the first transceiver.
12 . The cloud-based device according to claim 11 , wherein the output end of the first transceiver is connected to an input end of the first HMM analyzer and an input end of the voice signal processing unit.
13 . The cloud-based device according to claim 11 , wherein the output end of the first transceiver is connected to one or more of input ends of the scenario HMM analyzer, the audience expression HMM analyzer, and the audience voice HMM analyzer and an input end of the voice signal processing unit.
14 . An operating method of a cloud-based device, comprising:
receiving, by using a first transceiver in the device, input data coming from a second transceiver of a robot;
receiving an audience voice input signal from the first transceiver in the device by using a voice signal processing unit in an analysis apparatus in the device, identifying the audience voice input signal, and outputting a standard instruction based on an identification result;
respectively receiving, by using a first HMM analyzer in the analysis apparatus in the device, a scenario input signal, an audience expression input signal, and an audience voice input signal that come from the first transceiver in the device, and using them as observable sequences of a first HMM;
deducing, by the first HMM analyzer, a hidden state change sequence of the first HMM based on an observable sequence probability maximization criterion, and outputting the hidden state change sequence to an emotional state HMM analyzer in the analysis apparatus, wherein the hidden state change sequence of the first HMM comprises a scenario hidden state change sequence, an audience expression hidden state change sequence, and an audience voice hidden state change sequence;
receiving, by the emotional state HMM analyzer, the scenario hidden state change sequence, the audience expression hidden state change sequence, and the audience voice hidden state change sequence, using them as observable sequences of an emotional state HMM, and deducing a hidden state change sequence of the emotional state HMM based on the observable sequence probability maximization criterion; and
selecting, by a decision apparatus in the device, humorous behavior based on the hidden state change sequence of the emotional state HMM, and consolidating a humorous behavior instruction and the standard instruction to obtain a final output instruction.
15 . The method according to claim 14 , wherein the first HMM analyzer further comprises a scenario HMM analyzer, an audience expression HMM analyzer, and an audience voice HMM analyzer that are connected in a series or parallel manner, wherein the scenario HMM analyzer, the audience expression HMM analyzer, and the audience voice HMM analyzer respectively receive a scenario input signal, an audience expression input signal, and an audience voice input signal; use them as observable sequences of a scenario HMM, an audience expression HMM, and an audience voice HMM; deduce hidden state change sequences of the scenario HMM, the audience expression HMM, and the audience voice HMM based on the observable sequence probability maximization criterion; and send the hidden state change sequences of the scenario HMM, the audience expression HMM, and the audience voice HMM to the emotional state HMM analyzer.
16 . The method according to claim 14 , wherein the step of selecting, by a decision apparatus in the device, humorous behavior based on the hidden state change sequence of the emotional state HMM, and consolidating a humorous behavior instruction and the standard instruction to obtain a final output instruction comprises:
receiving, by a humorous behavior selection unit in the decision apparatus in the device, the hidden state change sequence of the emotional state HMM, performing probability analysis on the received hidden state change sequence of the emotional state HMM, selecting the humorous behavior, and outputting the humorous behavior instruction to a consolidation unit in the decision apparatus; and
receiving, by the consolidation unit, the humorous behavior instruction and the standard instruction, and consolidating the humorous behavior instruction and the standard instruction to obtain the final output instruction.
17 . The method according to claim 16 , wherein the consolidation further comprises:
when the humorous behavior instruction is one of “telling a joke”, “reading interesting news”, “performing a funny action”, and “singing”, the consolidation unit selects an optimal humor output instruction by searching a cloud database and/or accessing the Internet with reference to the audience voice input signal, and uses the optimal humor output instruction and the standard instruction as the final output instruction, wherein the optimal humor output instruction is an instruction most matching an emotional state of a target audience.
18 . The method according to claim 17 , wherein a related policy for selecting the humorous behavior and the optimal humor output instruction is correspondingly adjusted by using feedback information obtained based on continuous interaction with the target audience.
19 . The method according to claim 16 , wherein the probability analysis comprises: calculating, by the humorous behavior selection unit, humorous behavior set probability distribution by using a preset probability transition matrix from an emotional state to a humorous behavior set.
20 . The method according to claim 14 , wherein deducing the hidden state change sequence based on the observable sequence probability maximization criterion is implemented by using a Viterbi algorithm.