Intent recognition and emotional text-to-speech learning
An example intent-recognition system comprises a processor and memory storing instructions. The instructions cause the processor to receive speech input comprising spoken words. The instructions cause the processor to generate text results based on the speech input and generate acoustic feature annotations based on the speech input. The instructions also cause the processor to apply an intent model to the text result and the acoustic feature annotations to recognize an intent based on the speech input. An example system for adapting an emotional text-to-speech model comprises a processor and memory. The memory stores instructions that cause the processor to receive training examples comprising speech input and receive labelling data comprising emotion information associated with the speech input. The instructions also cause the processor to extract audio signal vectors from the training examples and generate an emotion-adapted voice font model based on the audio signal vectors and the labelling data.
1. An intent recognition system, the system comprising:
at least one processor; and
memory, operatively connected to the at least one processor and storing instructions that, when executed by the at least processor, cause the at least one processor to:
receive a speech input comprising spoken words;
generate a text result based on the speech input, wherein the text result comprises a textual representation of the spoken words in the speech input;
generate acoustic feature annotations based on the speech input, the acoustic feature annotations including at least spectral harmonics of the speech input;
determine an emotion associated with the speech input; and
apply an intent model to the text result and the acoustic feature annotations to recognize an intent based on the speech input, wherein, when the emotion is determined to be sarcasm, the recognized intent is to perform no action with respect to a device.
2. The system of claim 1 , wherein the acoustic feature annotations include annotations related to hyper-textual features of the speech input.
3. The system of claim 1 , wherein the acoustic feature annotations include annotations associated with the entire speech input.
4. The system of claim 1 , wherein the recognized intent comprises a tone associated with the speech input.
5. The system of claim 1 , wherein the intent model comprises a recursive neural network.
6. The system of claim 1 , wherein the memory further stores instructions that, when executed by the at least one processor, cause the at least one processor to:
perform an action based on the recognized intent; and
update a user interface based on the results of the action.
7. The system of claim 1 , wherein the memory further stores instructions that, when executed by the at least one processor, cause the at least one processor to:
generate first prosody annotations for the speech input;
generate second prosody annotations from labelling data associated with the speech input;
determine differences between the first prosody annotations and the second prosody annotations; and
generate an emotional text-to-speech model based on the determined differences between the first prosody annotations and the second prosody annotations.
8. The system of claim 1 , wherein applying the intent model comprises:
determine whether the intent is ambiguous;
when the intent is determined to be ambiguous, evaluate the text result and the acoustic feature annotations; and
when the intent is determined to be unambiguous, evaluate the text result and do not evaluate the acoustic feature annotations.
9. The system of claim 1 , wherein the intent model is:
a hidden Markov model;
a segmental model; or
a maximum entropy model.
10. The system of claim 1 , wherein the memory further stores instructions that, when executed by the at least one processor, cause the at least one processor to:
after recognizing the intent, providing a synthesized speech response having emotion that is based on the speech input.
11. A computer readable media not consisting of a propagated data signal, the computer readable media storing executable instructions that, when executed by a computer, cause the computer to perform operations comprising:
receiving a speech input comprising spoken words;
generating a text result based on the speech input, wherein the text result comprises a textual representation of the spoken words in the speech input;
generating acoustic feature annotations based on the speech input, the acoustic feature annotations including at least spectral harmonics of the speech input;
determining an emotion associated with the speech input; and
applying an intent model to the text result and the acoustic feature annotations to recognize an intent based on the speech input, wherein, when the emotion is determined to be sarcasm, the recognized intent is to perform no action with respect to a device.
12. The computer readable media of claim 11 , wherein the acoustic feature annotations include annotations related to hyper-textual features of the speech input.
13. The computer readable media of claim 11 , wherein the acoustic feature annotations include annotations associated with the entire speech input.
14. The computer readable media of claim 11 , wherein the recognized intent comprises a tone associated with the speech input.
15. The method of claim 11 , wherein the intent model comprises a recursive neural network.
16. The computer readable media of claim 11 , further comprising:
performing an action based on the recognized intent; and
updating a user interface based on the results of the action.
17. The computer readable media of claim 11 , further comprising:
generating first prosody annotations for the speech input;
generating second prosody annotations from labelling data associated with the speech input;
determining differences between the first prosody annotations and the second prosody annotations; and
generating an emotional text-to-speech model based on the determined differences between the first prosody annotations and the second prosody annotations.
18. The computer readable media of claim 11 , wherein applying the intent model comprises:
determining whether the intent is ambiguous;
when the intent is determined to be ambiguous, evaluating the text result and the acoustic feature annotations; and
when the intent is determined to be unambiguous, evaluating the text result and do not evaluate the acoustic feature annotations.
19. The computer readable media of claim 11 , further comprising:
after recognizing the intent, providing a synthesized speech response having emotion that is based on the speech input.
20. A device comprising:
a processing unit; and
a memory storing computer executable instructions that, when executed by the processing unit, causes the device to perform operations, comprising:
receiving a speech input comprising spoken words;
generating a text result based on the speech input, wherein the text result comprises a textual representation of the spoken words in the speech input;
generating acoustic feature annotations based on the speech input, the acoustic feature annotations including at least spectral harmonics of the speech input;
determining an emotion associated with the speech input; and
applying an intent model to the text result and the acoustic feature annotations to recognize an intent based on the speech input, wherein, when the emotion is determined to be sarcasm, the recognized intent is to perform no action with respect to a device.