Semantic recognition method and semantic recognition device

1. A semantic recognition method, comprising:

generating a spectrogram of a speech signal;

detecting at least one keyword of the spectrogram by inputting the spectrogram into a neural network model to detect at least one target feature block corresponding to the at least one keyword from the spectrogram, wherein the target feature block comprises part of feature points of the spectrogram, the neural network model output a classification category of the at least one target feature block within the spectrogram, and the at least one keyword is the classification category;

distinguishing a semantic category to which each of the at least one keyword belongs; and

determining a semantic intention of the speech signal according to the at least one keyword and the semantic category of the at least one keyword.

2. The semantic recognition method as claimed in claim 1 , wherein the step of generating the spectrogram of the speech signal comprises:

using fast Fourier transformation to convert the speech signal into the spectrogram, wherein the spectrogram comprises time information, frequency information, and intensity information of the speech signal.

3. The semantic recognition method as claimed in claim 1 , wherein the step of generating the spectrogram of the speech signal comprises:

using fast Fourier transformation and Mel-filtering to convert the speech signal into the spectrogram, wherein the spectrogram comprises time information, frequency information, and intensity information of the speech signal.

4. The semantic recognition method as claimed in claim 1 , wherein the neural network model comprises a convolution neural network model or a neural network model applying an attention mechanism.

5. The semantic recognition method as claimed in claim 1 , wherein the step of determining the semantic intention of the speech signal according to the at least one keyword and the semantic category of the at least one keyword comprises:

filling the at least one keyword into at least one semantic slot of a semantic frame according to the semantic category of the at least one keyword; and

determining the semantic intention according to the at least one keyword filled in the at least one semantic slot in response to all of the at least one semantic slot being filled.

6. The semantic recognition method as claimed in claim 5 , wherein the step of filling the at least one keyword into the at least one semantic slot of the semantic frame according to the semantic category of the at least one keyword comprises:

selecting to fill a first keyword having a first classification probability into the at least one semantic slot of the semantic frame when the semantic category of the first keyword in the at least one keyword is the same as the semantic category of a second keyword in the at least one keyword, wherein the first classification probability of the first keyword and a second classification probability of the second keyword are generated by the neural network model, and the first classification probability is greater than the second classification probability.

7. A semantic recognition method, comprising:

generating a spectrogram of a speech signal;

detecting at least one keyword of the spectrogram by inputting the spectrogram into a neural network model to detect at least one target feature block corresponding to the at least one keyword from the spectrogram, wherein the target feature block comprises part of feature points of the spectrogram, the neural network model is built up based on a relationship between each first keyword of the at least one keyword and the part of feature points of the spectrogram in the target feature block corresponding to the first keyword;

distinguishing a semantic category to which each of the at least one keyword belongs; and

determining a semantic intention of the speech signal according to the at least one keyword and the semantic category of the at least one keyword.

8. A semantic recognition device, comprising:

a storage device; and

a processing circuit, coupled to the storage device and configured to execute instructions in the storage device to:

generate a spectrogram of a speech signal;

detect at least one keyword of the spectrogram by inputting the spectrogram into a neural network model to detect at least one target feature block corresponding to the at least one keyword from the spectrogram, wherein the target feature block comprises part of feature points of the spectrogram, the neural network model output a classification category of the at least one target feature block within the spectrogram, and the at least one keyword is the classification category;

distinguish a semantic category to which each of the at least one keyword belongs; and

determine a semantic intention of the speech signal according to the at least one keyword and the semantic category of the at least one keyword.

9. The semantic recognition device as claimed in claim 8 , wherein the processing circuit is configured to:

use fast Fourier transformation to convert the speech signal into the spectrogram, wherein the spectrogram comprises time information, frequency information, and intensity information of the speech signal.

10. The semantic recognition device as claimed in claim 8 , wherein the processing circuit is configured to:

use fast Fourier transformation and Mel-filtering to convert the speech signal into the spectrogram, wherein the spectrogram comprises time information, frequency information, and intensity information of the speech signal.

11. The semantic recognition device as claimed in claim 8 , wherein the neural network model comprises a convolution neural network model or a neural network model applying an attention mechanism.

12. The semantic recognition device as claimed in claim 8 , wherein the processing circuit is configured to:

fill the at least one keyword into at least one semantic slot of a semantic fame according to the semantic category of the at least one keyword; and

determine the semantic intention according to the at least one keyword filled in the at least one semantic slot in response to all of the at least one semantic slot being filled.

13. The semantic recognition device as claimed in claim 12 , wherein the processing circuit is configured to:

select to fill a first keyword having a first classification probability into the at least one semantic slot of the semantic frame when the semantic category of the first keyword in the at least one keyword is the same as the semantic category of a second keyword in the at least one keyword, wherein the first classification probability of the first keyword and a second classification probability of the second keyword are generated by the neural network model, and the first classification probability is greater than the second classification probability.