A method and device for identifying intention and related equipment

Abstract

The present disclosure provides an intention recognition method, apparatus and related equipment, the method comprising: obtaining a phoneme sequence corresponding to the audio to be detected, and determining a text sequence corresponding to the audio to be detected based on the phoneme sequence; extracting and fusing the text sequence and the phoneme sequence based on a cross-attention mechanism to obtain a sentence feature vector corresponding to the audio to be detected; and obtaining a prediction result for characterizing the intention of the audio to be detected based on the sentence feature vector. The method can improve the accuracy of spoken intention recognition.

Claims

1. A method for identifying intent, It is characterized in that The method includes: Acquire a phoneme sequence corresponding to the audio to be detected, and determine a text sequence corresponding to the audio to be detected based on the phoneme sequence; Based on the cross attention mechanism, feature extraction and fusion are performed on the text sequence and the phoneme sequence to obtain a sentence feature vector corresponding to the audio to be detected; Based on the sentence feature vector, a prediction result for characterizing the intention of the audio to be detected is obtained, wherein the step of extracting and fusing features of the text sequence and the phoneme sequence based on a cross attention mechanism to obtain a sentence feature vector corresponding to the audio to be detected includes: Encoding the text sequence and the phoneme sequence respectively to obtain a text sequence feature vector corresponding to the text sequence and a phoneme sequence feature vector corresponding to the phoneme sequence; The text sequence feature vector and the phoneme sequence feature vector are concatenated according to the text sequence weight parameter and the phoneme sequence weight parameter to obtain the sentence feature vector, wherein the text sequence weight parameter and the phoneme sequence weight parameter are obtained by training a plurality of pre-stored audio samples.

2. The method for identifying intention according to claim 1, It is characterized in that The step of obtaining a prediction result for characterizing the intention of the audio to be detected based on the sentence feature vector specifically includes: The sentence feature vector is input into a fully connected neural network for classification prediction to obtain the prediction result.

3. The method for identifying intention according to claim 1, It is characterized in that The step of encoding the text sequence and the phoneme sequence to obtain a text sequence feature vector corresponding to the text sequence and a phoneme sequence feature vector corresponding to the phoneme sequence includes: A bidirectional neural network is used to encode the text sequence and the phoneme sequence to obtain a text sequence feature vector corresponding to the text sequence and a phoneme sequence feature vector corresponding to the phoneme sequence.

4. The method for identifying intention according to claim 1, It is characterized in that The text sequence weight parameter and the phoneme sequence weight parameter are obtained by training a plurality of pre-stored audio samples, specifically including: Encoding a sample text sequence and a sample phoneme sequence to obtain a sample text sequence feature vector corresponding to the sample text sequence and a sample phoneme sequence feature vector corresponding to the sample phoneme sequence; wherein the sample text sequence and the sample phoneme sequence are obtained by performing speech recognition on the audio sample; Calculating the cosine similarity between the embedding vector of each text in the sample text sequence feature vector and the sample phoneme sequence feature vector to obtain the weight of each text, and normalizing the weights corresponding to all texts through a normalized exponential function to obtain the text sequence weight parameter; The phoneme sequence weight parameter is obtained according to the text sequence weight parameter.

5. The method for identifying intention according to claim 4, It is characterized in that The text sequence weight parameter and the phoneme sequence weight parameter are adjusted in the following manner, specifically including: determining a true text transcribed from said audio sample listened to by a human; Based on a cross entropy loss function, calculating the difference between a prediction result representing the intent of the audio sample and the real text; Based on the difference, the text sequence weight parameter and the phoneme sequence weight parameter are adjusted.

6. An intention recognition device, It is characterized in that The device is used to implement the intention recognition method described in any one of claims 1 to 5, and the device includes: A speech recognition module, used to obtain a phoneme sequence corresponding to the audio to be detected, and determine a text sequence corresponding to the audio to be detected based on the phoneme sequence; A cross attention module, used for extracting and fusing features of the text sequence and the phoneme sequence based on a cross attention mechanism, and obtaining a sentence feature vector corresponding to the audio to be detected; A fully connected neural network classification module is used to obtain a prediction result for characterizing the intention of the audio to be detected based on the sentence feature vector.

7. The intention recognition device according to claim 6, It is characterized in that The fully connected neural network classification module includes: The sentence feature vector is input into a fully connected neural network for classification prediction to obtain the prediction result.

8. The intention recognition device according to claim 7, It is characterized in that The cross attention module specifically includes: An encoding unit encodes the text sequence and the phoneme sequence respectively to obtain a text sequence feature vector corresponding to the text sequence and a phoneme sequence feature vector corresponding to the phoneme sequence; A concatenation unit concatenates the text sequence feature vector and the phoneme sequence feature vector according to the text sequence weight parameter and the phoneme sequence weight parameter to obtain the sentence feature vector.

9. The intention recognition device according to claim 8, It is characterized in that The encoding unit specifically includes: A bidirectional neural network is used to encode the text sequence and the phoneme sequence to obtain a text sequence feature vector corresponding to the text sequence and a phoneme sequence feature vector corresponding to the phoneme sequence.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the program, the steps of the intention recognition method as described in any one of claims 1 to 5 are implemented.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the intention recognition method as described in any one of claims 1 to 5 are implemented.

Images