An AI conversational robot system based on voice information
By using feature words and cloud platform data determination mechanism in AI dialogue robot system, combining time intervals and feature word ratios, the problem of low quality testing efficiency in the existing technology is solved, the dialogue accuracy and user experience are improved, and the realization of the realization of the conversation is achieved. Dynamic adjustment of the system is used to improve overall performance.
Patent Information
- Application Number
- CN202410802864.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-20
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-06-20
AI Technical Summary
In the prior art, the hostage testing and testing of dialogue robots is inefficient, resulting in poor dialogue accuracy and user experience, and the inability to accurately identify user intentions, affecting the efficiency of subsequent AI use.
Design an AI dialogue robot system based on voice information, obtain feature words through the receiving, transforming, extracting and analyzing modules, combine cloud platform data to determine whether the session information meets the preset standards, and use time intervals and feature word overlap for preliminary and Secondary judgment, and adjust the domain label, feature word extraction amount and filtering magnification based on the judgment results to improve the accuracy of the conversation.
The accuracy and user experience of dialogue robot feedback are improved. Through the determination of time intervals and characteristic word ratios, the conversation information complies with preset standards, and the system parameters are adjusted to improve conversation accuracy and user experience.
Smart Images

Figure CN118506781B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of artificial intelligence technology, and in particular to an AI dialogue robot system based on voice information. Background Art
[0002] With the development of artificial intelligence technology, the application of conversational robots is becoming more and more extensive. By adopting advanced speech recognition technology and natural language processing technology, they can quickly complete multiple rounds of conversations, and can help companies solve a large amount of repetitive, mechanical and simple labor in the shortest time, and on this basis quickly improve the company's work efficiency and service quality.
[0003] However, before the conversational robot goes online, it is an essential step to conduct quality inspection and testing on it. In the existing technology, test cases are usually written based on manual experience, and then the conversational robot is tested according to the test cases. The operation is cumbersome and the quality inspection efficiency is low, resulting in the inability to accurately identify users in later applications, resulting in inefficient use of subsequent AI. Summary of the invention
[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an AI dialogue robot system based on voice information to improve dialogue accuracy and improve user experience.
[0005] The present invention solves the technical problem by adopting the following technical solutions:
[0006] An AI conversational robot system based on voice information, comprising:
[0007] A receiving module, used for receiving the user's voice information and recording it as initial voice information or secondary voice information;
[0008] A conversion module, connected to the receiving module, for converting the received voice information into text information, and the converted text information is recorded as initial text information or secondary text information;
[0009] An extraction module, connected to the conversion module, for obtaining feature words in the text information, where the obtained feature words are recorded as initial feature words or secondary feature words;
[0010] An analysis module, connected to the extraction module, for determining the intention of the voice information based on the feature words, and then obtaining the corresponding conversation information from the cloud platform;
[0011] A cloud platform, which is connected to the analysis module and is used to store big data;
[0012] an output module, connected to the analysis module, for outputting session information;
[0013] The feedback module is connected to the extraction module and the analysis module respectively, and is used to detect the characteristic words of the secondary voice information to determine whether the output initial conversation information meets the preset standard.
[0014] Furthermore, the feedback module makes a preliminary judgment on whether the output initial conversation information meets the standard based on the time interval between the output initial conversation information and the user's secondary voice information. If the time interval is greater than the interval standard, it is preliminarily judged that the output initial conversation information does not meet the preset standard, and it is necessary to compare the overlap degree of the user's initial feature words and secondary feature words, otherwise it is not necessary.
[0015] Furthermore, if it is preliminarily determined that the outputted initial session information does not meet the preset standard, then the outputted initial session information is finally determined to meet the preset standard according to the overlap between the user's initial feature words and the secondary feature words; wherein the analysis module can also be used to determine whether the initial session information meets the preset standard;
[0016] If the overlap is less than the first overlap standard, the analysis module determines that the outputted initial conversation information meets the preset standard;
[0017] If the overlap is greater than or equal to the first overlap standard and less than the second overlap standard, the analysis module performs a secondary determination on the initial session information;
[0018] If the overlap is greater than or equal to the second overlap standard, the analysis module determines that the outputted initial session information does not meet the preset standard and determines the reason;
[0019] The first coincidence standard is smaller than the second coincidence standard, and the coincidence degree = the number of repeated words between the initial feature words and the secondary feature words / the total number of secondary feature words.
[0020] Furthermore, if a secondary determination is required for the output initial conversation information, the analysis module determines whether the output initial conversation information is qualified according to the ratio of the total number of secondary feature words to the total number of initial feature words;
[0021] If the word count ratio is greater than the ratio standard, the analysis module determines that the outputted initial conversation information is qualified, and then performs subsequent conversation feedback;
[0022] If the word count ratio is less than or equal to the ratio standard, the analysis module determines that the output initial conversation information is unqualified and determines the reason.
[0023] Further, if the word number ratio is greater than the ratio standard, the analysis module determines the adjustment range of the first coincidence degree and the second coincidence degree according to the ratio difference between the word number ratio and the ratio standard;
[0024] If the ratio difference is greater than the first preset difference, the analysis module determines that the reduction range of the first overlap degree and the second overlap degree is the first reduction range;
[0025] If the ratio difference is less than or equal to the first preset difference, the analysis module determines that the reduction range of the first overlap degree and the second overlap degree is the second reduction range;
[0026] Among them, the first reduction range is greater than the second reduction range.
[0027] Furthermore, if it is determined that the outputted initial session information does not meet the preset standard, the analysis module determines the reason according to the difference between the coincidence degree and the coincidence degree of the second coincidence degree standard.
[0028] If the overlap difference is less than the first preset overlap difference, the analysis module determines that there is a problem in the analysis process and needs to expand the domain label range of the feature word of the analysis module;
[0029] If the overlap difference is greater than or equal to the first preset overlap difference and less than the second preset overlap difference, the analysis module determines that the extraction amount of the feature words needs to be expanded;
[0030] If the overlap degree is greater than or equal to the second preset overlap degree difference, the analysis module determines that the filter magnification of the voice information extracted by the conversion module needs to be adjusted;
[0031] The first preset overlap difference is smaller than the second preset overlap difference.
[0032] Furthermore, if it is necessary to expand the domain label range of the feature word of the analysis module, the analysis module determines the number of selected domain labels according to the secondary difference, where the secondary difference is the difference between the overlap difference and the first preset overlap difference.
[0033] If the secondary difference is less than or equal to the first preset secondary difference, the analysis module determines that the selected number of domain labels is the first number;
[0034] If the secondary difference is greater than the first preset secondary difference and less than or equal to the second preset secondary difference, the analysis module determines that the number of selected domain labels is the second number;
[0035] If the secondary difference is greater than the second preset secondary difference, the analysis module determines that the selected number of field labels is a third number;
[0036] Among them, the first preset secondary difference is smaller than the second preset secondary difference, and the first number is larger than the second number and larger than the third number.
[0037] Furthermore, if it is necessary to expand the number of feature words to be extracted, the analysis module determines the adjustment coefficient for the initial number of feature words according to the length of the received text information.
[0038] If the length of the received text message is the first length, the analysis module determines that the adjustment coefficient for the number of initial feature words is the first adjustment coefficient;
[0039] If the length of the received text message is the second length, the analysis module determines that the adjustment coefficient for the number of initial feature words is the second adjustment coefficient;
[0040] If the length of the received text message is the third length, the analysis module determines that the adjustment coefficient for the number of initial feature words is the third adjustment coefficient;
[0041] Among them, the first length is smaller than the second length and smaller than the third length, and the first adjustment coefficient is smaller than the second adjustment coefficient and smaller than the third adjustment coefficient.
[0042] Furthermore, if it is necessary to adjust the filter magnification of the voice information extracted by the conversion module, the analysis module determines the correction coefficient of the filter magnification according to the time taken to convert the voice information into text information.
[0043] If the time taken to convert the voice information into text information is the first time, the analysis module determines the correction coefficient for the filtering multiple as the first correction coefficient;
[0044] If the time taken to convert the voice information into text information is the second time, the analysis module determines the correction coefficient for the filtering multiple as the second correction coefficient;
[0045] If the time taken to convert the voice information into text information is the third time, the analysis module determines the correction coefficient for the filtering multiple to be the third correction coefficient;
[0046] Among them, the first time is smaller than the second time and smaller than the third time, and the first correction coefficient is smaller than the second correction coefficient and smaller than the third correction coefficient.
[0047] The advantages and positive effects of the present invention are:
[0048] The present invention determines whether the output conversation information meets the preset standard through the time interval between the output initial conversation information and the user's secondary voice information and the ratio of the user's initial feature words and secondary feature words. If the time interval is smaller, it indicates that the user did not think about the subsequent conversation and directly repeated the previous conversation, and it is preliminarily determined that the output conversation information does not meet the standard; and if the repetition is higher, it indicates that the user has asked repeated questions, indirectly indicating that the conversation information output by the robot is unqualified; then, the output initial conversation information is determined to be qualified through the ratio of the total number of secondary feature words to the total number of initial feature words for a secondary judgment, and further determines whether this feedback solves the problem compared with the initial voice information, thereby improving the accuracy of the robot's feedback.
[0049] The present invention determines whether the output initial conversation information is qualified for secondary judgment based on the ratio of the total number of secondary feature words to the total number of initial feature words. If the ratio is larger, it indicates that this feedback has solved most of the problems and subsequent analysis and feedback can be carried out. If the ratio is smaller, it indicates that this feedback has not solved any problems. The specific reason is determined by the difference between the ratio and the second preset difference, and the robot system is adjusted according to the situation, thereby improving the accuracy of robot conversation and improving user experience.
[0050] The present invention is provided with an adjustment method for the number of field label selections, an adjustment method for the number of feature word extractions and an adjustment method for the filter ratio, so as to achieve adjustment of the robot system in different situations. Compared with the prior art, the robot system is no longer adjusted roughly, which further improves the accuracy of the robot dialogue and improves the user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION
[0052] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.
[0053] An AI conversational robot system based on voice information, comprising:
[0054] A receiving module, used for receiving the user's voice information and recording it as initial voice information or secondary voice information;
[0055] A conversion module, connected to the receiving module, for converting the received voice information into text information, and the converted text information is recorded as initial text information or secondary text information;
[0056] An extraction module, connected to the conversion module, for obtaining feature words in the text information, where the obtained feature words are recorded as initial feature words or secondary feature words;
[0057] An analysis module, connected to the extraction module, for determining the intention of the voice information based on the feature words, and then obtaining the corresponding conversation information from the cloud platform;
[0058] A cloud platform, which is connected to the analysis module and is used to store big data;
[0059] an output module, connected to the analysis module, for outputting session information;
[0060] The feedback module is connected to the extraction module and the analysis module respectively, and is used to detect the characteristic words of the secondary voice information to determine whether the output initial conversation information meets the preset standard.
[0061] The feedback module makes a preliminary judgment on whether the output initial conversation information meets the standard based on the time interval between the output initial conversation information and the user's secondary voice information. If the time interval is greater than the interval standard, it is preliminarily determined that the output initial conversation information does not meet the preset standard, and it is necessary to compare the overlap degree of the user's initial feature words and secondary feature words, otherwise it is not necessary.
[0062] In an embodiment of the present invention, the feedback module makes a preliminary judgment on whether the output initial conversation information meets the standard based on the time interval between the output initial conversation information and the user's secondary voice information. Assume that the time interval is S and the interval standard is S1. If S≤S1, the analysis module preliminarily determines that the output initial conversation information meets the preset standard; if S﹥S1, the analysis module preliminarily determines that the output initial conversation information does not meet the preset standard.
[0063] If it is preliminarily determined that the output initial conversation information does not meet the preset standard, the overlap between the user's initial feature words and the secondary feature words is set to C, C=Cn / Cz, Cn represents the number of repeated words between the initial feature words and the secondary feature words, Cz' represents the total number of secondary feature words, and the first overlap standard is set to C1, the second overlap standard is set to C2, and C1﹤C2; if C﹤C1, the analysis module determines that the output initial conversation information meets the preset standard; if C1≤C﹤C2, the analysis module performs a secondary judgment on the initial conversation information; if C≥C2, the analysis module determines that the output initial conversation information does not meet the preset standard and determines the reason.
[0064] If it is necessary to make a secondary judgment on the output initial conversation information, the analysis module determines whether the output initial conversation information is qualified according to the word number ratio of the total number of secondary feature words to the total number of initial feature words; the word number ratio is set to B, B=Cz' / Cz, where Cz represents the total number of initial feature words, and the ratio standard is set to B1. If B>B1, the analysis module determines that the output initial conversation information is qualified, and then performs subsequent conversation feedback; if B≤B1, the analysis module determines that the output initial conversation information is unqualified and determines the reason.
[0065] If the word count ratio is greater than the ratio standard, the adjustment range of the first overlap and the second overlap is determined according to the ratio difference Bc between the word count ratio B and the ratio standard B1, that is, Bc=|B-B1|, the first preset difference is set to Bc1, the first reduction range is D1, the second reduction range is D2, and D1>D2; if Bc>Bc1, the analysis module determines that the reduction range of the first overlap and the second overlap is the first reduction range; if Bc≤Bc1, the analysis module determines that the reduction range of the first overlap and the second overlap is the second reduction range.
[0066] If it is determined that the output initial conversation information does not meet the preset standard, the reason is determined according to the overlap difference Cc between the overlap C and the second overlap standard C2, that is, Cc=|C-C2|, the first preset overlap difference is set to Cc1, the second preset overlap difference is set to Cc2, and Cc1﹤Cc2; if Cc﹤Cc1, the analysis module determines that there is a problem in the analysis process, and it is necessary to expand the analysis module's domain label range for feature words; if Cc1≥Cc﹤Cc2, the analysis module determines that it is necessary to expand the extraction amount of feature words; if Cc≥Cc2, the analysis module determines that it is necessary to adjust the filtering ratio of the voice information extracted by the conversion module.
[0067] If it is necessary to expand the scope of domain labels for feature words of the analysis module, the number of selected domain labels is determined according to the secondary difference, the secondary difference is the difference between the overlap difference Cc and the first preset overlap difference Cc1, that is, A=|Cc-Cc1|, the first preset secondary difference is set to A1, the second preset secondary difference is set to A2, and A1﹤A2, the first number is set to L1, the second number is set to L2, the third number is set to L3, and L1﹥L2﹥L3; if A≤A1, the analysis module determines that the number of selected domain labels is the first number L1; if A1﹤A≤A2, the analysis module determines that the number of selected domain labels is the second number L2; if A﹥A2, the analysis module determines that the number of selected domain labels is the third number L3.
[0068] If it is necessary to expand the extraction amount of feature words, select an adjustment coefficient according to the length G of the received text information to adjust the initial number of feature words, set the length of the text information to the first length G1, the second length G2, and the third length G3, and G1 < G2 < G3, the adjustment coefficients are x1, x2, and x3, and 0.5 < x1 < x2 < x3 < 1, set the initial number of feature words to Y0, if the length of the received text information is G1, the analysis module determines to select the first adjustment coefficient x1 to adjust the number of initial feature words to the corresponding value; if the length of the received text information is G2, the analysis module determines to select the second adjustment coefficient x2 to adjust the number of initial feature words to the corresponding value; if the length of the received text information is G3, the analysis module determines to select the third adjustment coefficient x3 to adjust the number of initial feature words to the corresponding value;
[0069] If the i-th adjustment coefficient xi is used to adjust the number of initial feature words Y0 to a corresponding value, and i=1, 2, 3 is set, the number of feature words after adjustment is Yj=Y0×xi.
[0070] If it is necessary to adjust the filter magnification of the voice information extracted by the conversion module, select a correction coefficient according to the time taken for the voice information to be converted into text information to determine the adjustment of the filter magnification, set the information used to convert the voice information into text to be T1, T2, and T3, respectively, and T1 < T2 < T3, set the correction coefficients to be k1, k2, and k3, respectively, and 0.5 < k1 < k2 < k3 < 1, and set the initial filter magnification to Q0; if the time taken for the voice information to be converted into text information is T1, the analysis module selects the first correction coefficient k1 to correct the initial filter magnification to the corresponding value; if the time taken for the voice information to be converted into text information is T2, the analysis module selects the second correction coefficient k2 to correct the initial filter magnification to the corresponding value; if the time taken for the voice information to be converted into text information is T3, the analysis module selects the first correction coefficient k3 to correct the initial filter magnification to the corresponding value;
[0071] If the wth correction coefficient wi is used to correct the number of initial feature words Q0 to a corresponding value, and w=1, 2, 3 is set, the corrected filtering factor is Qv=Q0×wi.
[0072] Although the embodiments of the present invention are disclosed for illustrative purposes, those skilled in the art will appreciate that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments.
Claims
1. An AI conversational robot system based on voice information, characterized in that: include: A receiving module, used for receiving the user's voice information and recording it as initial voice information and secondary voice information; A conversion module, connected to the receiving module, for converting the received voice information into text information, and the converted text information is recorded as initial text information and secondary text information; An extraction module, connected to the conversion module, for obtaining feature words in the text information, wherein the obtained feature words are recorded as initial feature words and secondary feature words; An analysis module, connected to the extraction module, for determining the intention of the voice information based on the feature words, and then obtaining the corresponding conversation information from the cloud platform; A cloud platform, which is connected to the analysis module and is used to store big data; an output module, connected to the analysis module, for outputting session information; A feedback module, which is connected to the extraction module and the analysis module respectively, and is used to detect the characteristic words of the secondary voice information to determine whether the output initial conversation information meets the preset standard; The feedback module makes a preliminary judgment on whether the output initial conversation information meets the standard based on the time interval between the output initial conversation information and the user's secondary voice information. If the time interval is greater than the interval standard, it is preliminarily determined that the output initial conversation information does not meet the preset standard, and it is necessary to compare the overlap degree of the user's initial feature words and secondary feature words, otherwise it is not necessary.
2. The AI conversational robot system based on voice information according to claim 1, characterized in that: If it is preliminarily determined that the outputted initial session information does not meet the preset standard, then the outputted initial session information is finally determined to meet the preset standard according to the overlap between the user's initial feature words and the secondary feature words; wherein the analysis module is also used to determine whether the initial session information meets the preset standard; If the overlap is less than the first overlap standard, the analysis module determines that the outputted initial conversation information meets the preset standard; If the overlap is greater than or equal to the first overlap standard and less than the second overlap standard, the analysis module performs a secondary determination on the initial session information; If the overlap is greater than or equal to the second overlap standard, the analysis module determines that the outputted initial session information does not meet the preset standard and determines the reason; The first coincidence standard is smaller than the second coincidence standard, and the coincidence degree = the number of repeated words between the initial feature words and the secondary feature words / the total number of secondary feature words.
3. The AI conversational robot system based on voice information according to claim 2, characterized in that: If a secondary judgment is required for the output initial conversation information, the analysis module determines whether the output initial conversation information is qualified according to the ratio of the total number of secondary feature words to the total number of initial feature words; If the word count ratio is greater than the ratio standard, the analysis module determines that the outputted initial conversation information is qualified, and then performs subsequent conversation feedback; If the word count ratio is less than or equal to the ratio standard, the analysis module determines that the output initial conversation information is unqualified and determines the reason.
4. The AI conversational robot system based on voice information according to claim 3, characterized in that: If the word number ratio is greater than the ratio standard, the analysis module determines the adjustment range of the first coincidence degree and the second coincidence degree according to the ratio difference between the word number ratio and the ratio standard; If the ratio difference is greater than the first preset difference, the analysis module determines that the reduction range of the first overlap degree and the second overlap degree is the first reduction range; If the ratio difference is less than or equal to the first preset difference, the analysis module determines that the reduction range of the first overlap degree and the second overlap degree is the second reduction range; Among them, the first reduction range is greater than the second reduction range.
5. The AI conversational robot system based on voice information according to claim 2, characterized in that: If the output initial session information is determined not to meet the preset standard, the analysis module determines the reason according to the difference between the overlap degree and the second overlap degree standard. If the overlap difference is less than the first preset overlap difference, the analysis module determines that there is a problem in the analysis process and needs to expand the domain label range of the feature word of the analysis module; If the overlap difference is greater than or equal to the first preset overlap difference and less than the second preset overlap difference, the analysis module determines that the extraction amount of the feature words needs to be expanded; If the overlap degree is greater than or equal to the second preset overlap degree difference, the analysis module determines that the filter magnification of the voice information extracted by the conversion module needs to be adjusted; The first preset overlap difference is smaller than the second preset overlap difference.
6. The AI conversational robot system based on voice information according to claim 5, characterized in that: If it is necessary to expand the domain label range of the feature word of the analysis module, the analysis module determines the number of selected domain labels according to the secondary difference, where the secondary difference is the difference between the overlap difference and the first preset overlap difference. If the secondary difference is less than or equal to the first preset secondary difference, the analysis module determines that the selected number of domain labels is the first number; If the secondary difference is greater than the first preset secondary difference and less than or equal to the second preset secondary difference, the analysis module determines that the number of selected domain labels is the second number; If the secondary difference is greater than the second preset secondary difference, the analysis module determines that the selected number of field labels is a third number; Among them, the first preset secondary difference is smaller than the second preset secondary difference, and the first number is larger than the second number and larger than the third number.
7. The AI conversational robot system based on voice information according to claim 5, characterized in that: If it is necessary to expand the extraction quantity of feature words, the analysis module determines the adjustment coefficient of the initial number of feature words according to the length of the received text information. If the length of the received text message is the first length, the analysis module determines that the adjustment coefficient for the number of initial feature words is the first adjustment coefficient; If the length of the received text message is the second length, the analysis module determines that the adjustment coefficient for the number of initial feature words is the second adjustment coefficient; If the length of the received text message is the third length, the analysis module determines that the adjustment coefficient for the number of initial feature words is the third adjustment coefficient; Among them, the first length is smaller than the second length and smaller than the third length, and the first adjustment coefficient is smaller than the second adjustment coefficient and smaller than the third adjustment coefficient.
8. The AI conversational robot system based on voice information according to claim 5, characterized in that: If it is necessary to adjust the filter magnification of the voice information extracted by the conversion module, the analysis module determines the correction coefficient of the filter magnification according to the time taken to convert the voice information into text information. If the time taken to convert the voice information into text information is the first time, the analysis module determines the correction coefficient for the filtering multiple as the first correction coefficient; If the time taken to convert the voice information into text information is the second time, the analysis module determines the correction coefficient for the filtering multiple as the second correction coefficient; If the time taken to convert the voice information into text information is the third time, the analysis module determines the correction coefficient for the filtering multiple to be the third correction coefficient; Among them, the first time is smaller than the second time and smaller than the third time, and the first correction coefficient is smaller than the second correction coefficient and smaller than the third correction coefficient.
Citation Information
Patent Citations
Voice recognition system based on Internet of Things
CN111785278A
Voice interactive device
JP2001022384A
Method of detecting a transcription error in speech recognition corpus and device for the same
US20240194203A1
Speech recognition result evaluation method, apparatus and device, and storage medium
WO2022095353A1
Method for testing speech interaction system, audio recognition method, and related devices
WO2022199461A1
