Emotion Analysis Method Based on User Voice Information

Abstract

The present invention provides an emotion analysis method based on user voice information, including: obtaining the voice information of an analysis object in real time, determining a voice text and additional voice information according to the voice information; generating emotion auxiliary information based on the additional voice information, extracting emotional vocabulary from the voice text according to the emotion auxiliary information, and determining the emotion analysis result of the user when the voice text is issued according to the text emotional vocabulary.

Description

Technical Field

[0001] The present invention relates to the technical field of emotion recognition, and in particular, to an emotion analysis method based on user voice information. Background Art

[0002] According to the "Report on the Development of Chinese National Mental Health (2017-2018)" jointly released by the Institute of Psychology of the Chinese Academy of Sciences and the Social Sciences Academic Press, the current methods for analyzing emotions online in China are mainly through questionnaire surveys, which have large errors and obvious subjectivity, and poor accuracy; moreover, an individual's cognition is affected by many aspects and the accuracy is not high.

[0003] In the current field of intelligent customer service, the current method for recognizing emotions in speech mainly involves extracting audio feature vectors of the recognized speech segments and matching them with multiple emotion feature models, and classifying the emotion corresponding to the emotion feature model with a matching result as the emotion classification of the speech segment. This method has wide recognition but lacks pertinence and has relatively large errors. The existing technology mainly recognizes emotions by obtaining audio feature vectors in the acquired audio stream and then matching the audio feature vectors with emotion feature models. This method has certain limitations. For example, it ignores characteristics such as the speech rate, intonation, and personal language habits in the speech segment, and the accuracy of emotion recognition is not high. Therefore, there is an urgent need for an emotion analysis method based on user voice information to solve the above problems. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides an emotion analysis method based on user voice information to solve the above problems.

[0005] An emotion analysis method based on user voice information includes: obtaining the voice information of the analysis object in real time, and determining the speech text and additional voice information according to the voice information;

[0006] Generating emotion auxiliary information based on the additional voice information, extracting emotional vocabulary from the speech text according to the emotion auxiliary information, and determining the emotion analysis result of the user when the speech text is issued according to the text emotional vocabulary.

[0007] As an embodiment of the present invention, an emotion analysis method based on user voice information further includes:

[0008] Sorting the speech texts according to the acquisition time of the voice information to generate a sorting result n, where n is the number of speech texts in the sorting result;

[0009] When n is 1, generate first emotion assistance information based on the first additional voice information, extract emotional vocabulary from the first voice text according to the first emotion assistance information, and determine the first emotion analysis result of the user when the voice text is issued according to the first text emotional vocabulary;

[0010] When n is 2, generate second emotion assistance information based on the second additional voice information, extract emotional vocabulary from the second voice text according to the second emotion assistance information, and determine the second emotion analysis result of the user when the voice text is issued according to the second text emotional vocabulary and the first emotion analysis result;

[0011] When n is m, generate the m-th emotion assistance information based on the m-th additional voice information, extract emotional vocabulary from the m-th voice text according to the m-th emotion assistance information, and determine the m-th emotion analysis result of the user when the voice text is issued according to the m-th text emotional vocabulary and the (m - 1)-th emotion analysis result, where m is an integer greater than 2.

[0012] As an embodiment of the present invention, determining the m-th emotion analysis result of the user when the voice text is issued according to the m-th text emotional vocabulary and the (m - 1)-th emotion analysis result includes:

[0013] Obtain the m-th text emotional vocabulary, and determine the initial emotion analysis result corresponding to the m-th text emotional vocabulary according to the preset text emotional vocabulary - emotion comparison table;

[0014] Judge whether the emotion fluctuation between the initial emotion analysis result and the (m - 1)-th emotion analysis result conforms to the preset range. If it conforms, use the initial emotion analysis result as the m-th emotion analysis result;

[0015] If it does not conform, determine an emotion grading strategy according to the emotion categories of the initial emotion analysis result and the (m - 1)-th emotion analysis result, and mark the initial emotion analysis result according to the emotion grading strategy to generate the m-th emotion analysis result.

[0016] As an embodiment of the present invention, determining the voice text and the additional voice information according to the voice information includes:

[0017] Collect voice information and convert it into text to determine the voice text;

[0018] Extract voice feature information in the voice information to determine the additional voice information; wherein, the voice feature information includes: voice line feature information, speech rate feature information, semantic feature information, and intonation amplitude change feature information.

[0019] As an embodiment of the present invention, generating emotion assistance information based on the additional voice information includes:

[0020] Obtain additional voice information, associate the additional voice information with each voice text, and determine the emotion auxiliary information of each voice text.

[0021] As an embodiment of the present invention, extracting emotional words from the voice text according to the emotion auxiliary information includes:

[0022] Obtain the voice text, and extract the words with emotions in the voice text as the initial emotional words;

[0023] Obtain the meanings of all the initial emotional words, and generate text emotional words according to the meanings of all the initial emotional words and the semantic feature information of the current voice text.

[0024] As an embodiment of the present invention, determining the emotion analysis result of the user when the voice text is issued according to the text emotional words includes:

[0025] Obtain the text emotional words, and determine the emotion analysis result of the user when the voice text is issued according to the preset text emotional word - emotion comparison table.

[0026] As an embodiment of the present invention, when obtaining the meanings of all the initial emotional words and generating text emotional words according to the meanings of all the initial emotional words and the semantic feature information of the current voice text, it includes:

[0027] Obtain the meanings of all the initial emotional words, perform meaning matching on each initial emotional word according to the preset meaning matching database, and determine the standard word of each initial emotional word;

[0028] Obtain the semantic feature information of the current voice text, perform semantic matching on the semantic feature information according to the preset semantic matching database, and determine the semantic word corresponding to the current voice text;

[0029] Calculate the semantic similarity between the standard word of each initial emotional word and the semantic word, and statistically calculate the average semantic similarity between the standard words of all the initial emotional words and the semantic word;

[0030] Perform meaning matching on the preset word matching database according to the average semantic similarity and the semantic word to determine the text emotional words.

[0031] As an embodiment of the present invention, a method for emotion analysis based on user voice information further includes:

[0032] Judge whether there is any semantic similarity between the meaning of an initial emotional word and the text emotional word less than the preset similarity;

[0033] If so, extract the corresponding initial emotional word as the reverse word to be predicted;

[0034] Obtain the voice feature information, speech rate feature information, and intonation amplitude change feature information corresponding to the reverse vocabulary to be predicted as prediction auxiliary information, and judge whether there is an error in the meaning of the reverse vocabulary to be predicted according to the prediction auxiliary information. If there is an error, generate a correction value;

[0035] Correct the meaning of the reverse vocabulary to be predicted according to the correction value to generate a corrected meaning;

[0036] Replace the meaning of the reverse vocabulary to be predicted in all initial emotion words according to the corrected meaning, and generate text emotion words according to the corrected meanings of all initial emotion words and the semantic feature information of the current speech text.

[0037] As an embodiment of the present invention, a method for emotion analysis based on user voice information further includes:

[0038] Obtain all initial emotion words, and judge whether there is an error in the meaning of the corresponding initial emotion word based on the prediction auxiliary information of each initial emotion word. If there is an error, generate a second correction value;

[0039] Correct the meaning of the initial emotion word with an error according to the second correction value to generate a second corrected meaning;

[0040] Replace the meaning of the corresponding initial emotion word with an error according to the second corrected meaning;

[0041] After the correction is completed, obtain the meanings of all initial emotion words and the semantic feature information of the current speech text to generate text emotion words.

[0042] The beneficial effects of the present invention are:

[0043] The present invention provides a method for emotion analysis based on user voice information, which is beneficial for voice customer service to be able to distinguish the user's emotion in a timely manner when facing the user, so as to formulate a more appropriate response strategy to meet the user's needs and improve user satisfaction.

[0044] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification and the drawings.

[0045] The following will further describe the technical solutions of the present invention in detail through the drawings and embodiments. Description of the Drawings

[0046] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0047] Figure 1 This is the method flowchart of an emotion analysis method based on user voice information in an embodiment of the present invention;

[0048] Figure 2 This is the generation flowchart of text emotion vocabulary in an emotion analysis method based on user voice information in an embodiment of the present invention;

[0049] Figure 3 This is another method flowchart in an emotion analysis method based on user voice information in an embodiment of the present invention. Detailed implementation manners

[0050] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0051] Please refer to Figure 1 , an emotion analysis method based on user voice information, including:

[0052] S101. Real-time obtain the voice information of the analysis object, and determine the voice text and additional voice information according to the voice information;

[0053] S102. Generate emotion auxiliary information based on the additional voice information, extract emotion vocabulary from the voice text according to the emotion auxiliary information, and determine the emotion analysis result of the user when the voice text is sent according to the text emotion vocabulary;

[0054] The working principle of the above technical solution is as follows: This application provides an emotion analysis method based on user voice information, preferably applied to the customer service call scenario; First, real-time obtain the voice information of the analysis object, determine the voice text and additional voice information according to the voice information, then generate emotion auxiliary information based on the additional voice information, extract emotion vocabulary from the voice text according to the emotion auxiliary information, and determine the emotion analysis result of the user when the voice text is sent according to the text emotion vocabulary;

[0055] The beneficial effect of the above technical solution is: Through the above technical solution, it is beneficial for the voice customer service to be able to distinguish the user's emotion in a timely manner when facing the user, so as to formulate a more appropriate reply strategy to meet the user's needs and improve the user satisfaction.

[0056] In one embodiment, an emotion analysis method based on user voice information further includes:

[0057] Sort the voice text according to the acquisition time of the voice information to generate a sorting result n, where n is the number of voice texts in the sorting result;

[0058] When n is 1, generate first emotion assistance information based on the first additional voice information, extract emotional vocabulary from the first voice text according to the first emotion assistance information, and determine the first emotion analysis result of the user when the voice text is issued according to the first text emotional vocabulary;

[0059] When n is 2, generate second emotion assistance information based on the second additional voice information, extract emotional vocabulary from the second voice text according to the second emotion assistance information, and determine the second emotion analysis result of the user when the voice text is issued according to the second text emotional vocabulary and the first emotion analysis result;

[0060] When n is m, generate the m-th emotion assistance information based on the m-th additional voice information, extract emotional vocabulary from the m-th voice text according to the m-th emotion assistance information, and determine the m-th emotion analysis result of the user when the voice text is issued according to the m-th text emotional vocabulary and the (m - 1)-th emotion analysis result, where m is an integer greater than 2;

[0061] The beneficial effects of the above technical solution are as follows: Through the above technical solution, by diversifying the acquisition of analysis samples, the context connection between texts is improved, and the accuracy of emotion result analysis is strengthened.

[0062] In one embodiment, determining the m-th emotion analysis result of the user when the voice text is issued according to the m-th text emotional vocabulary and the (m - 1)-th emotion analysis result includes:

[0063] Obtain the m-th text emotional vocabulary, and determine the initial emotion analysis result corresponding to the m-th text emotional vocabulary according to the preset text emotional vocabulary - emotion comparison table;

[0064] Judge whether the emotion fluctuation between the initial emotion analysis result and the (m - 1)-th emotion analysis result conforms to the preset range. If it conforms, use the initial emotion analysis result as the m-th emotion analysis result;

[0065] If it does not conform, determine an emotion grading strategy according to the emotion categories of the initial emotion analysis result and the (m - 1)-th emotion analysis result, and mark the initial emotion analysis result according to the emotion grading strategy to generate the m-th emotion analysis result;

[0066] The working principle and beneficial effects of the above technical solution are as follows: Obtain the m-th text emotional vocabulary, and determine the initial emotion analysis result corresponding to the m-th text emotional vocabulary according to the preset text emotional vocabulary - emotion comparison table; Determine whether the emotion fluctuation between the initial emotion analysis result and the (m - 1)-th emotion analysis result conforms to the preset range. If it conforms, use the initial emotion analysis result as the m-th emotion analysis result; If not, determine the emotion grading strategy according to the emotion categories of the initial emotion analysis result and the (m - 1)-th emotion analysis result, mark the initial emotion analysis result according to the emotion grading strategy, and generate the m-th emotion analysis result. Through the above technical solution, the context connection between texts is improved, and the analysis accuracy of emotion results is enhanced.

[0067] In one embodiment, determining the speech text and additional speech information according to the speech information includes:

[0068] Collect the speech information and convert it into text to determine the speech text;

[0069] Extract the speech feature information in the speech information to determine the additional speech information; wherein, the speech feature information includes: voice line feature information, speech rate feature information, semantic feature information, and intonation amplitude change feature information;

[0070] The beneficial effect of the above technical solution is: Through the above technical solution, diverse information is extracted from the speech information, providing more comprehensive and reliable basic data for subsequent emotion analysis.

[0071] In one embodiment, generating emotion auxiliary information based on the additional speech information includes:

[0072] Obtain the additional speech information, associate the additional speech information with each speech text, and determine the emotion auxiliary information of each speech text;

[0073] The working principle and beneficial effects of the above technical solution are as follows: The association relationship between each speech text and the additional speech information is determined by each speech information. Preferably, when processing the speech information, the corresponding speech text and the additional speech information are associated. Data classification is beneficial to accelerating the subsequent analysis and processing efficiency.

[0074] Please refer to Figure 2 , in one embodiment, extracting emotional vocabulary from the speech text according to the emotion auxiliary information includes:

[0075] S201. Obtain the speech text, and extract the words containing emotions in the speech text as the initial emotional vocabulary;

[0076] S202. Obtain the meanings of all the initial emotional vocabulary, and generate text emotional vocabulary according to the meanings of all the initial emotional vocabulary and the semantic feature information of the current speech text;

[0077] The working principle and beneficial effects of the above technical solution are as follows: Obtain the speech text, extract the words with emotions in the speech text as the initial emotion words, obtain the meanings of all the initial emotion words, generate text emotion words according to the meanings of all the initial emotion words and the semantic feature information of the current speech text, and by jointly analyzing the semantics of the overall text and the semantics of individual emotion words, it is beneficial to improve the accuracy of emotion result analysis.

[0078] In one embodiment, determining the emotion analysis result of the user when the speech text is issued according to the text emotion words includes:

[0079] Obtain the text emotion words, and determine the emotion analysis result of the user when the speech text is issued according to the preset text emotion word - emotion comparison table.

[0080] In one embodiment, obtaining the meanings of all the initial emotion words and generating text emotion words according to the meanings of all the initial emotion words and the semantic feature information of the current speech text includes:

[0081] Obtain the meanings of all the initial emotion words, perform meaning matching on each initial emotion word according to the preset meaning matching database, and determine the standard word of each initial emotion word;

[0082] Obtain the semantic feature information of the current speech text, perform semantic matching on the semantic feature information according to the preset semantic matching database, and determine the semantic word corresponding to the current speech text;

[0083] Calculate the semantic similarity between the standard word of each initial emotion word and the semantic word, and statistically calculate the average semantic similarity between the standard words of all the initial emotion words and the semantic word;

[0084] Perform meaning matching on the preset word matching database according to the average semantic similarity and the semantic word to determine the text emotion words;

[0085] The working principle and beneficial effects of the above technical solution are as follows: Obtain the meanings of all the initial emotion words, perform meaning matching on each initial emotion word according to the preset meaning matching database, and determine the standard word of each initial emotion word; Obtain the semantic feature information of the current speech text, perform semantic matching on the semantic feature information according to the preset semantic matching database, and determine the semantic word corresponding to the current speech text; Calculate the semantic similarity between the standard word of each initial emotion word and the semantic word, and statistically calculate the average semantic similarity between the standard words of all the initial emotion words and the semantic word; Perform meaning matching on the preset word matching database according to the average semantic similarity and the semantic word to determine the text emotion words. Through the above technical solution, the text emotion words are determined, providing reliable data support for subsequent emotion analysis.

[0086] Please refer to Figure 3 , in one embodiment, an emotion analysis method based on user voice information further includes:

[0087] S301. Determine whether the semantic similarity between the semantic meaning of any initial emotion word and the semantic meaning of the text emotion word is less than a preset similarity;

[0088] 302. If so, extract the corresponding initial emotion word as the reverse word to be predicted;

[0089] 303. Obtain the voice line feature information, speech rate feature information, and intonation amplitude change feature information corresponding to the reverse word to be predicted as prediction auxiliary information, and determine whether there is an error in the semantic meaning of the reverse word to be predicted according to the prediction auxiliary information. If there is an error, generate a correction value;

[0090] 304. Correct the semantic meaning of the reverse word to be predicted according to the correction value to generate a corrected semantic meaning;

[0091] 305. Replace the semantic meaning of the reverse word to be predicted in all initial emotion words according to the corrected semantic meaning, and generate text emotion words according to the semantic feature information of all initial emotion words after correction and the current voice text;

[0092] The working principle and beneficial effects of the above technical solution are as follows: Determine whether the semantic similarity between the semantic meaning of any initial emotion word and the semantic meaning of the text emotion word is less than a preset similarity; if so, extract the corresponding initial emotion word as the reverse word to be predicted; obtain the voice line feature information, speech rate feature information, and intonation amplitude change feature information corresponding to the reverse word to be predicted as prediction auxiliary information, and determine whether there is an error in the semantic meaning of the reverse word to be predicted according to the prediction auxiliary information. If there is an error, generate a correction value; correct the semantic meaning of the reverse word to be predicted according to the correction value to generate a corrected semantic meaning; replace the semantic meaning of the reverse word to be predicted in all initial emotion words according to the corrected semantic meaning, and generate text emotion words according to the semantic feature information of all initial emotion words after correction and the current voice text. Through the above technical solution, the acquisition breadth and depth of text emotion words are improved, and the accuracy of emotion recognition is improved.

[0093] In one embodiment, an emotion analysis method based on user voice information further includes:

[0094] Obtain all initial emotion words, and determine whether there is an error in the semantic meaning of the corresponding initial emotion word based on the prediction auxiliary information of each initial emotion word. If there is an error, generate a second correction value;

[0095] Correct the semantic meaning of the initial emotion word with an error according to the second correction value to generate a second corrected semantic meaning;

[0096] Replace the meaning of the initial sentiment word with corresponding errors according to the second corrected meaning.

[0097] After the correction is completed, obtain the meanings of all initial sentiment words and the semantic feature information of the current speech text to generate text sentiment words.

[0098] The working principle and beneficial effects of the above technical solution are as follows: Obtain all initial sentiment words, judge whether there are errors in the meanings of the corresponding initial sentiment words based on the prediction auxiliary information of each initial sentiment word. If there are errors, generate a second correction value; correct the meanings of the initial sentiment words with corresponding errors according to the second correction value to generate second corrected meanings; replace the meanings of the corresponding initial sentiment words with errors according to the second corrected meanings; after the correction is completed, obtain the meanings of all initial sentiment words and the semantic feature information of the current speech text to generate text sentiment words. Through the above technical solution, the acquisition breadth and depth of text sentiment words are improved, and the accuracy of emotion recognition is improved.

[0099] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for emotion analysis based on user voice information, characterized in that, Including: Obtaining the voice information of the analysis object in real time, and determining the voice text and additional voice information according to the voice information; Generating emotion auxiliary information based on the additional voice information, extracting emotional words from the voice text according to the emotion auxiliary information, and determining the emotion analysis result of the user when the voice text is issued according to the text emotional words; Extracting emotional words from the voice text according to the emotion auxiliary information, including: Obtaining the voice text, and extracting the words with emotions in the voice text as the initial emotional words; Obtaining the meanings of all the initial emotional words, and generating text emotional words according to the meanings of all the initial emotional words and the semantic feature information of the current voice text; Also including: Judging whether the semantic similarity between the meaning of any initial emotional word and the meaning of the text emotional word is less than the preset similarity; If so, extracting the corresponding initial emotional word as the reverse word to be predicted; Obtaining the voice line feature information, speech rate feature information and intonation amplitude change feature information corresponding to the reverse word to be predicted as the prediction auxiliary information, and judging whether there is an error in the meaning of the reverse word to be predicted according to the prediction auxiliary information. If there is an error, generating a correction value; Correcting the meaning of the reverse word to be predicted according to the correction value to generate a corrected meaning; Replacing the meaning of the reverse word to be predicted in all the initial emotional words with the corrected meaning, and generating text emotional words according to the corrected meanings of all the initial emotional words and the semantic feature information of the current voice text.

2. The emotional analysis method based on user voice information according to claim 1, wherein, Also including: Sorting the voice texts according to the acquisition time of the voice information to generate a sorting result n, where n is the number of voice texts in the sorting result; When n is 1, generating the first emotion auxiliary information based on the first additional voice information, extracting emotional words from the first voice text according to the first emotion auxiliary information, and determining the first emotion analysis result of the user when the voice text is issued according to the first text emotional words; When n is 2, generating the second emotion auxiliary information based on the second additional voice information, extracting emotional words from the second voice text according to the second emotion auxiliary information, and determining the second emotion analysis result of the user when the voice text is issued according to the second text emotional words and the first emotion analysis result; When n is m, generating the mth emotion auxiliary information based on the mth additional voice information, extracting emotional words from the mth voice text according to the mth emotion auxiliary information, and determining the mth emotion analysis result of the user when the voice text is issued according to the mth text emotional words and the (m - 1)th emotion analysis result, where m is an integer greater than 2.

3. The method for emotion analysis based on user voice information according to claim 2, wherein, Determining the mth emotion analysis result of the user when the voice text is issued according to the mth text emotional words and the (m - 1)th emotion analysis result, including: Obtaining the mth text emotional words, and determining the initial emotion analysis result corresponding to the mth text emotional words according to the preset text emotional word - emotion comparison table; Judging whether the emotion fluctuation between the initial emotion analysis result and the (m - 1)th emotion analysis result conforms to the preset range. If it conforms, using the initial emotion analysis result as the mth emotion analysis result; If not, determine an emotion classification strategy according to the emotion categories of the initial emotion analysis result and the (m-1)th emotion analysis result, and mark the initial emotion analysis result according to the emotion classification strategy to generate the mth emotion analysis result.

4. The emotional analysis method based on user voice information according to claim 1, wherein Determine the speech text and additional speech information according to the speech information, including: Collect speech information and convert it into text to determine the speech text; Extract the speech feature information in the speech information to determine the additional speech information; wherein, the speech feature information includes: voice line feature information, speech rate feature information, semantic feature information, and intonation amplitude change feature information.

5. A method for emotion analysis based on user voice information according to claim 1, characterized in that, Generate emotion auxiliary information based on the additional speech information, including: Obtain the additional speech information, associate the additional speech information with each speech text, and determine the emotion auxiliary information of each speech text.

6. The emotional analysis method based on user voice information according to claim 1 is characterized in that, Determine the emotion analysis result of the user when the speech text is issued according to the text emotion vocabulary, including: Obtain the text emotion vocabulary, and determine the emotion analysis result of the user when the speech text is issued according to the preset text emotion vocabulary-emotion comparison table.

7. A method for emotion analysis based on user voice information according to claim 1, characterized in that, It also includes: Obtain all initial emotion vocabulary, judge whether there is an error in the meaning of the corresponding initial emotion vocabulary based on the prediction auxiliary information of each initial emotion vocabulary, and if there is an error, generate a second correction value; Correct the meaning of the initial emotion vocabulary with an error according to the second correction value to generate a second corrected meaning; Replace the meaning of the corresponding initial emotion vocabulary with an error with the second corrected meaning; After the correction is completed, obtain the meaning of all initial emotion vocabulary and the semantic feature information of the current speech text to generate text emotion vocabulary.

Images