Intelligent management method of enterprise data based on speech analysis

By dividing enterprise meeting audio into local audio segments and differentiating them according to keywords and the importance of the speaker, and using different compression algorithms for transmission, the problems of high resource consumption and low efficiency in enterprise meeting audio transmission are solved, achieving efficient transmission and the integrity of important information.

CN118737165BActive Publication Date: 2024-11-08李刚
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411203575.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-08
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

In existing technologies, transmitting corporate meeting audio to a corporate database using lossless compression algorithms consumes a lot of resources and time, making it impossible to efficiently store corporate meeting audio.

Method used

The audio of the enterprise meeting is divided into local audio segments of preset duration. Based on the distribution of meeting keywords and the importance of the speakers within the local audio segments, key local audio segments and non-key local audio segments are determined. The key local audio segments are transmitted using a lossless compression algorithm, while the non-key local audio segments are transmitted using a lossy compression algorithm.

Benefits of technology

It improves the efficiency of audio transmission in enterprise meetings, saves resources, and ensures the integrity and accuracy of important voice information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118737165B_ABST
    Figure CN118737165B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of speech recognition technology, and specifically to an enterprise data intelligent management method based on speech analysis. The method obtains enterprise conference audio and divides it into local audio segments; obtains the criticality of the local audio segments according to the distribution of conference keywords in the local audio segments; obtains the speaker's audio importance according to the distribution of conference conjunctions and conference keywords in the speaker's corresponding audio segments; corrects the criticality according to the speaker's audio duration and audio importance in the local audio segments, obtains the actual criticality, determines the critical local audio segments and non-critical local audio segments, and compresses and transmits the enterprise conference audio. The present invention determines the critical local audio segments and non-critical local audio segments, and then compresses and transmits the enterprise conference audio separately, so that the accuracy of the key information of the enterprise meeting is ensured during the transmission process, and the efficiency of the enterprise conference audio transmission is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to an enterprise data intelligent management method based on speech analysis. Background Art

[0002] In order to record the content of each company meeting, it is necessary to record the content of each company meeting and then transfer the meeting audio to the company database for storage.

[0003] The existing method uses a lossless compression algorithm to transmit the conference audio to the enterprise database to ensure the audio quality. However, in large enterprises, corporate meetings are frequent and the audio of each meeting is long. Transmitting the conference audio to the enterprise database through a lossless compression algorithm consumes a lot of resources and time, and it is impossible to store the corporate conference audio efficiently. Summary of the invention

[0004] In order to solve the technical problem that transmitting conference audio to the enterprise database through a lossless compression algorithm consumes a lot of resources and time and cannot efficiently store the enterprise conference audio, the purpose of the present invention is to provide an enterprise data intelligent management method based on speech analysis. The technical solution adopted is as follows:

[0005] An embodiment of the present invention provides an enterprise data intelligent management method based on speech analysis, the method comprising the following steps:

[0006] Obtain enterprise conference audio, and divide the enterprise conference audio into local audio segments of preset duration;

[0007] According to the distribution of conference keywords in each local audio segment, the criticality of each local audio segment is obtained;

[0008] According to the distribution of conference connecting words and conference keywords in the corresponding audio segments of each speaker in each enterprise meeting, the audio importance of each speaker in each enterprise meeting is obtained;

[0009] According to the audio duration and audio importance of each speaker in each local audio segment, the criticality of each local audio segment is corrected to obtain the actual criticality of each local audio segment;

[0010] Based on the actual criticality, critical local audio segments and non-critical local audio segments are determined, and compression transmission management is performed on the critical local audio segments and the non-critical local audio segments respectively.

[0011] Furthermore, the method for obtaining the criticality is:

[0012] For any local audio segment, the ratio of the number of conference keywords appearing in the local audio segment to the number of all constituent words in the local audio segment is used as the keyword probability of the local audio segment;

[0013] The ratio of the number of conference keyword types appearing in the local audio segment to the number of all conference keyword types is used as the keyword correlation value of the local audio segment;

[0014] The criticality of the local audio segment is obtained according to the keyword probability and the keyword relevance value of the local audio segment; wherein both the keyword probability and the keyword relevance value are positively correlated with the criticality.

[0015] Furthermore, the method for obtaining the audio importance is:

[0016] For any speaker of any corporate meeting, the audio importance of the speaker is obtained based on the keyword relevance value of the speaker's corresponding audio segment and the number of occurrences of the conference connecting words in the speaker's corresponding audio segment; wherein, the keyword relevance value of the speaker's corresponding audio segment and the number of occurrences of the conference connecting words are both positively correlated with the audio importance.

[0017] Furthermore, the method for obtaining the actual criticality is:

[0018] For any local audio segment, the proportion of the audio duration of each speaker in the local audio segment in the local audio segment is used as the first feature value of each speaker;

[0019] Obtaining an important weight of each speaker in the local audio segment according to the first eigenvalue and the audio importance of each speaker in the local audio segment; wherein the first eigenvalue and the audio importance are both positively correlated with the important weight;

[0020] According to the importance weight of each speaker in the local audio segment and the criticality of the local audio segment, the actual criticality of the local audio segment is obtained; wherein the importance weight and the criticality are both positively correlated with the actual criticality.

[0021] Furthermore, the method for obtaining the key local audio segment and the non-key local audio segment is:

[0022] When the actual criticality meets the preset criticality range, the corresponding local audio segment is used as the critical local audio segment;

[0023] When the actual criticality does not meet the preset criticality range, the corresponding local audio segment is regarded as a non-critical local audio segment.

[0024] Furthermore, the enterprise data intelligent management method based on speech analysis also includes:

[0025] For any key local audio segment, when a local audio segment adjacent to the key local audio segment is a non-key local audio segment, any non-key local audio segment adjacent to the key local audio segment is used as a reference audio segment;

[0026] Obtain a speaker whose audio is present in both the key local audio segment and the reference audio segment as a reference speaker;

[0027] Obtaining a second feature value of each reference speaker according to the audio duration of each reference speaker in the key local audio segment and the reference audio segment;

[0028] According to the second eigenvalue of each reference speaker and the actual criticality difference between the key local audio segment and the reference audio segment, the coherence degree of the key local audio segment and the reference audio segment is obtained; wherein the second eigenvalue is positively correlated with the coherence degree, and the actual criticality difference is negatively correlated with the coherence degree;

[0029] When the coherence level meets a preset coherence level range, the reference audio segment is modified into a key local audio segment.

[0030] Furthermore, the method for obtaining the second characteristic value is:

[0031] For any reference speaker, obtain the cumulative duration of all audio durations of the reference speaker in the key local audio segment and the reference audio segment as the first duration;

[0032] The cumulative result of the corresponding durations of the key local audio segment and the reference audio segment is used as the second duration;

[0033] The ratio of the first duration to the second duration is used as the second characteristic value of the reference speaker.

[0034] Furthermore, the method for respectively performing compression transmission management on the key local audio segment and the non-key local audio segment is:

[0035] The key local audio segment is transmitted using a lossless compression algorithm;

[0036] The non-critical local audio segments are transmitted using a lossy compression algorithm.

[0037] Furthermore, the method for obtaining the conference keywords and conference connecting words is:

[0038] By manually setting the meeting keywords and meeting connecting words for each corporate meeting audio; among them, the meeting connecting words include summary words and emergency words.

[0039] Furthermore, the constituent words in the enterprise meeting audio are obtained through the Stanford CoreNLP natural speech processing tool.

[0040] The present invention has the following beneficial effects:

[0041] Dividing the corporate conference audio into local audio segments of preset duration is conducive to more complete transmission and preservation of important voice information in the corporate conference audio; then, based on the distribution of conference keywords in each local audio segment, the criticality of each local audio segment is obtained, the importance of each local audio segment is preliminarily judged, and the compression of each local audio segment during transmission is predicted; in order to more accurately analyze the importance of each local audio segment, the audio importance of each speaker in each corporate meeting is obtained based on the distribution of conference conjunctions and conference keywords in the corresponding audio segment of each speaker in each corporate meeting, and the importance of each speaker's speech in the corporate meeting is determined, which is conducive to subsequent accurate modification of the criticality of each local audio segment. Accurately analyze the importance of the voice information in each local audio segment; then correct the criticality of each local audio segment according to the audio duration and audio importance of each speaker in each local audio segment, obtain the actual criticality of each local audio segment, accurately determine the importance of the voice information in each local audio segment, and then accurately determine the key local audio segments and non-key local audio segments based on the actual criticality, compress and transmit the key local audio segments separately, compress and transmit the key local audio segments losslessly, and compress and transmit the non-key local audio segments lossily, so as to improve the efficiency of enterprise conference audio transmission, save resources for transmitting enterprise conference audio, and ensure the integrity and accuracy of important voice information in enterprise conference audio. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic flow chart of an enterprise data intelligent management method based on speech analysis provided by one embodiment of the present invention;

[0044] Figure 2 A flow chart of a method for obtaining a criticality provided by an embodiment of the present invention;

[0045] Figure 3A flow chart of a method for obtaining actual criticality provided by an embodiment of the present invention;

[0046] Figure 4 A structural diagram of an enterprise data intelligent management system based on speech analysis provided by one embodiment of the present invention;

[0047] Figure 5 A schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the enterprise data intelligent management method based on speech analysis proposed by the present invention, its specific implementation method, structure, features and effects are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0049] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0050] The specific scheme of the enterprise data intelligent management method based on speech analysis provided by the present invention is described in detail below with reference to the accompanying drawings.

[0051] Embodiment 1:

[0052] The present invention proposes an enterprise data intelligent management method based on speech analysis, please refer to Figure 1 , which shows a schematic flow chart of an enterprise data intelligent management method based on speech analysis provided by an embodiment of the present invention, the method comprising the following steps:

[0053] Step S1: Acquire the enterprise conference audio, and divide the enterprise conference audio into local audio segments of preset duration.

[0054] Specifically, this embodiment takes any meeting of any enterprise as an example for analysis. It should be noted that the subsequent enterprise meetings all refer to the meeting of the enterprise selected in this embodiment. Before the enterprise meeting starts, a recorder is accurately set up, and the recorder is placed in the conference room. When the enterprise meeting starts, the recorder is turned on to record the enterprise meeting and obtain the enterprise meeting audio. If the enterprise meeting lasts for a long time, the enterprise meeting audio obtained is too long, and a lot of resources and time will be consumed in the process of losslessly compressing the enterprise meeting audio and transmitting it to the enterprise database. It is known that in actual situations, not all discussions during the enterprise meeting are important. Therefore, this embodiment divides the enterprise meeting audio into a plurality of local audio segments of preset duration, and then analyzes each local audio segment to determine the importance of each local audio segment, and transmits the important local audio segment losslessly, and transmits the unimportant local audio segment losslessly, while improving the efficiency of transmitting the enterprise meeting audio to the enterprise database, ensuring that the important content in the enterprise meeting is accurately transmitted to the enterprise database. Among them, this embodiment sets the preset duration to 10 seconds, and the implementer can set the size of the preset duration according to the actual situation, which is not limited here. It should be noted that when the last audio segment of the enterprise conference audio is less than 10 seconds, it is also defaulted to a partial audio segment.

[0055] Step S2: According to the distribution of conference keywords in each local audio segment, the criticality of each local audio segment is obtained.

[0056] Specifically, in actual situations, the content of each corporate meeting is different, that is, the meeting keywords in each corporate meeting audio are different. Therefore, this embodiment manually sets the meeting keywords in the corporate meeting audio. Among them, when the number and types of meeting keywords in a certain local audio segment are more, it means that the local audio segment is more critical, the voice information in the local audio segment is more important, and the local audio segment should be losslessly compressed and transmitted to the corporate database. Therefore, this embodiment obtains the criticality of each local audio segment based on the distribution of meeting keywords in each local audio segment. Among them, the greater the criticality, the more important the voice information in the corresponding local audio segment.

[0057] Preferably, in some possible implementations of this embodiment, the method for obtaining the criticality is described in Figure 2 , which shows a flow chart of a method for obtaining a criticality provided by an embodiment of the present invention, the method comprising:

[0058] Step S201: For any partial audio segment, the ratio of the number of conference keywords appearing in the partial audio segment to the number of all constituent words in the partial audio segment is used as the keyword probability of the partial audio segment.

[0059] This embodiment uses the Stanford CoreNLP natural speech processing tool to obtain various words such as verbs and nouns in the enterprise conference audio, all of which are used as constituent words of the enterprise conference audio. Conference keywords also belong to constituent words. Among them, the Stanford CoreNLP natural speech processing tool is a well-known technology and will not be described in detail. The more conference keywords appear in a certain local audio segment, the more important the voice information in the local audio segment may be. Therefore, for any local audio segment, this embodiment uses the ratio of the number of conference keywords appearing in the local audio segment to the number of all constituent words in the local audio segment as the keyword probability of the local audio segment, wherein the larger the keyword probability, the more conference keywords appear in the local audio segment, and the more important the local audio segment is.

[0060] Step S202: taking the ratio of the number of conference keyword types appearing in the partial audio segment to the number of all conference keyword types as the keyword relevance value of the partial audio segment.

[0061] When the number of conference keywords appearing in a certain local audio segment is large, but the types of conference keywords are small, it means that the coverage of the conference keywords involved in the local audio segment is relatively narrow, which indirectly means that the less key information involved in the voice information in the local audio segment, the lower the importance of the local audio segment. Therefore, this embodiment further obtains the ratio of the number of types of conference keywords appearing in a certain local audio segment to the number of all types of conference keywords as the keyword relevance value of the local audio segment, wherein the larger the keyword relevance value, the more types of conference keywords appearing in the local audio segment, that is, the more key conference information involved in the local audio segment, the more important the local audio segment. It should be noted that when the number of conference keywords appearing in a certain local audio segment is 0, the keyword relevance value of the local audio segment is defaulted to 0.

[0062] Step S203: obtaining the criticality of the partial audio segment according to the keyword probability and the keyword relevance value of the partial audio segment; wherein both the keyword probability and the keyword relevance value are positively correlated with the criticality.

[0063] It is known that when the keyword probability and the keyword relevance value are both larger, it means that the voice information in the corresponding local audio segment is more critical. Therefore, this embodiment obtains the criticality of each local audio segment through the keyword probability and the keyword relevance value of each local audio segment; wherein the keyword probability and the keyword relevance value are both positively correlated with the criticality. Among them, the greater the criticality, the more important the voice information in the corresponding local audio segment. In order to more accurately reflect the importance of each local audio segment, the criticality of each local audio segment is further specifically quantified to obtain the criticality value of each local audio segment, wherein the greater the criticality value, the more important the voice information in the corresponding local audio segment. Among them, the expression of the criticality value is: ; In the formula, is the critical value of the vth local audio segment; is the number of all words in the vth local audio segment; The number of all conference keyword types that appear in the enterprise conference audio; is the keyword probability of the vth local audio segment; is the number of conference keyword types appearing in the vth local audio segment; is the keyword correlation value of the vth local audio segment. In other embodiments, the criticality value can be obtained by adding the keyword probability and the keyword correlation value, and the method for obtaining the criticality value is not limited here. Among them, the larger the criticality value, the more the corresponding local audio segment should be losslessly compressed and transmitted to the enterprise database to ensure that the voice information in the corresponding local audio segment is accurately stored.

[0064] Step S3: Obtain the audio importance of each speaker in each enterprise meeting according to the distribution of meeting conjunctions and meeting keywords in the audio segment corresponding to each speaker in each enterprise meeting.

[0065] Specifically, during a corporate meeting, speakers of different positions give speeches, and the content of speeches by different speakers is different. Therefore, the importance of the audio corresponding to each speaker is also different. The more important the speech content of a speaker is, the more critical the audio of the speaker is. Therefore, this embodiment analyzes the audio segment corresponding to each speaker in the corporate meeting to determine the importance of the audio of each speaker in the corporate meeting. The greater the importance of the audio, the more important the audio of the corresponding speaker is, and the more it should be losslessly compressed and transmitted to the corporate database, which indirectly indicates that the local audio segment where the audio of the corresponding speaker is located should be losslessly compressed and transmitted to the corporate database. Therefore, this embodiment obtains the audio importance of each speaker in the corporate meeting, further corrects the criticality of each local audio segment, and more accurately analyzes the importance of each local audio segment.

[0066] After performing voice recognition on the audio of the enterprise meeting, the audio corresponding to each speaker in the enterprise meeting can be obtained. In the enterprise meeting, when the speaker summarizes the enterprise meeting, some iconic words are usually used, such as summary words such as first, second, first, next and last. The summary words indicate that the speaker is sorting out the content discussed in the enterprise meeting in order, and the key content of the meeting is often involved in the local audio segment where the summary words are located. Therefore, when there are more summary words in the corresponding audio segment of a certain speaker, it means that the importance of the corresponding audio segment of the speaker is higher. At the same time, when there are more urgent words such as immediate and urgent need in the corresponding audio segment of a certain speaker, it means that the speech content of the speaker may involve an emergency or an important decision, and the speech content of the speaker may have high urgency, which indirectly indicates that the importance of the corresponding audio segment of the speaker is relatively high. Among them, this embodiment artificially sets the summary words and emergency words in the enterprise meeting audio, and in order to better describe it, the summary words and emergency words in the enterprise meeting audio are unified as conference connection words. When the types of conference connection words and conference keywords in the corresponding audio segment of a certain speaker are more, it means that the voice information in the corresponding audio segment of the speaker is more important. Therefore, this embodiment obtains the audio importance of each speaker in each enterprise meeting according to the distribution of conference conjunctions and conference keywords in the audio segment corresponding to each speaker in each enterprise meeting.

[0067] Preferably, in some possible implementations of this embodiment, the method for obtaining the audio importance is: for any speaker of any enterprise meeting, the audio importance of the speaker is obtained according to the keyword correlation value of the speaker's corresponding audio segment and the number of occurrences of the conference connection words in the speaker's corresponding audio segment. It should be noted that the speaker's corresponding audio segment is the sum of all the corresponding audios of the speaker's speech during the enterprise meeting. Among them, the method for obtaining the keyword correlation value of the speaker's corresponding audio segment can refer to the method for obtaining the keyword correlation value of the local audio segment in step S2, which will not be repeated here. When the keyword correlation value of the speaker's corresponding audio segment is larger, it means that the more types of conference keywords appear in the speaker's corresponding audio segment, the more important the speaker's speech content is, and the greater the audio importance of the speaker. At the same time, when the number of occurrences of the conference connection words in the speaker's corresponding audio segment is more, it indirectly means that the more key meeting information in the speaker's corresponding audio segment is, the greater the audio importance of the speaker. Therefore, the keyword correlation value of the speaker's corresponding audio segment and the number of occurrences of the conference connection words are both positively correlated with the audio importance. In order to accurately represent the importance of each speaker's speech at the enterprise meeting, this embodiment specifically quantifies the audio importance of each speaker and obtains the audio importance value of each speaker, where the larger the audio importance value, the more important the conference speech content of the corresponding speaker. The expression of the audio importance value is: ; In the formula, is the audio importance value of the i-th speaker; is the number of occurrences of the conference connective in the audio segment corresponding to the i-th speaker; The number of all conference keyword types that appear in the enterprise conference audio; is the number of conference keyword types that appear in the audio segment corresponding to the i-th speaker; is the keyword-related value of the audio segment corresponding to the i-th speaker; is the normalized exponential function.

[0068] Step S4: According to the audio duration and audio importance of each speaker in each partial audio segment, the criticality of each partial audio segment is corrected to obtain the actual criticality of each partial audio segment.

[0069] Specifically, in the process of discussing issues in a corporate meeting, there will be audios of multiple speakers in a local audio segment, but there will be no overlap in the audios of different speakers. In order to more accurately determine the importance of each local audio segment, this embodiment obtains the audio duration of each speaker in each local audio segment. When the audio importance of a certain speaker is greater and the audio duration in a certain local audio segment is longer, it can be said that the voice information in the local audio segment is more important. Therefore, this embodiment corrects the criticality of each local audio segment according to the audio duration and audio importance of each speaker in each local audio segment, and obtains the actual criticality of each local audio segment. Among them, the greater the actual criticality, the more important the voice information in the corresponding local audio segment, and the more the corresponding local audio segment should be losslessly compressed and transmitted to the enterprise database to ensure that the voice information represented by the corresponding local audio segment is accurately saved.

[0070] Preferably, in some possible implementations of this embodiment, the method for obtaining the actual criticality is described in Figure 3 , which shows a flow chart of a method for obtaining an actual criticality provided by an embodiment of the present invention, the method comprising:

[0071] Step S401: For any local audio segment, the proportion of the audio duration of each speaker in the local audio segment in the local audio segment is used as the first feature value of each speaker.

[0072] When the first eigenvalue is larger, the more the audio of the corresponding speaker is in the corresponding local audio segment, the more the importance of the audio of the corresponding speaker can reflect the importance of the corresponding local audio segment. Therefore, when correcting the criticality of the corresponding local audio segment, the probability of referring to the importance of the audio of the corresponding speaker is greater. The expression of the first eigenvalue is: ; In the formula, is the first eigenvalue of the r-th speaker in the a-th local audio segment; is the duration corresponding to the ath local audio segment; is the audio duration of the rth speaker in the ath local audio segment.

[0073] Step S402: obtaining an important weight of each speaker in the local audio segment according to the first eigenvalue and the audio importance of each speaker in the local audio segment; wherein the first eigenvalue and the audio importance are both positively correlated with the important weight.

[0074] For a certain speaker in a certain local audio segment, when the first eigenvalue and audio importance of the speaker are both larger, it means that the speaker's audio can better reflect the characteristics of the local audio segment, and thus the speaker's importance weight should be larger to ensure accurate correction of the criticality of the local audio segment. Therefore, this embodiment obtains the importance weight of each speaker in a certain local audio segment based on the first eigenvalue and audio importance of each speaker in the local audio segment; wherein the first eigenvalue and audio importance are both positively correlated with the importance weight. wherein, the larger the importance weight, the more accurately the audio importance of the corresponding speaker can reflect the importance of the local audio segment. wherein, the expression of the importance weight is: ; In the formula, is the importance weight of the r-th speaker in the a-th local audio segment; is the first eigenvalue of the r-th speaker in the a-th local audio segment; is the audio importance value of the rth speaker. In other embodiments, the importance weight can be obtained by adding the first feature value to the audio importance value, and the method for obtaining the importance weight is not limited here.

[0075] Step S403: obtaining the actual criticality of the local audio segment according to the importance weight of each speaker in the local audio segment and the criticality of the local audio segment; wherein both the importance weight and the criticality are positively correlated with the actual criticality.

[0076] By obtaining the important weight of each speaker in each local audio segment, and then according to the important weight of each speaker in each local audio segment, the criticality of each local audio segment is accurately modified, and the actual criticality of each local audio segment is accurately obtained. Among them, the larger the important weight, the more important the corresponding local audio segment is indirectly indicated. Therefore, the important weight and the criticality are both positively correlated with the actual criticality. In this embodiment, the result of accumulating the important weight of each speaker in each local audio segment is used as the overall weight of each local audio segment, and then according to the overall weight and criticality of each local audio segment, the actual criticality of each local audio segment is obtained, wherein the overall weight and the criticality are both positively correlated with the actual criticality. Among them, the larger the actual criticality, the more important the corresponding local audio segment is, and the more it should be losslessly compressed and transmitted to the enterprise database. In order to more accurately represent the importance of each local audio segment, this embodiment specifically quantifies the actual criticality of each local audio segment, and obtains the actual criticality value of each local audio segment, wherein the larger the actual criticality value, the more the corresponding local audio segment should be losslessly compressed and transmitted to the enterprise database. Among them, the expression of the actual critical value is: ; In the formula, is the actual criticality value of the vth local audio segment; is the total number of speakers in the vth local audio segment; is the importance weight of the jth speaker in the vth local audio segment; is the overall weight; is the critical value of the vth local audio segment.

[0077] Step S5: determining the critical local audio segments and the non-critical local audio segments based on the actual criticality, and performing compression transmission management on the critical local audio segments and the non-critical local audio segments respectively.

[0078] Specifically, it is known that the larger the actual criticality value is, the more important the corresponding local audio segment is. Therefore, in this embodiment, the actual criticality value of each local audio segment is normalized to obtain the normalized actual criticality value of each local audio segment. When the normalized actual criticality value is greater than or equal to the preset actual criticality value threshold, the corresponding local audio segment is regarded as the key local audio segment, and the key local audio segment is losslessly compressed and transmitted to the enterprise database to ensure the integrity of the key local audio segment; when the normalized actual criticality value is less than the preset actual criticality value threshold, the corresponding local audio segment is regarded as the non-key local audio segment, and the non-key local audio segment is losslessly compressed and transmitted to the enterprise database, thereby improving the efficiency of transmitting the enterprise conference audio to the enterprise database and saving resources for transmitting the enterprise conference audio. In this embodiment, the preset actual criticality value threshold is set to 0.8. The implementer can set the size of the preset actual criticality value threshold according to the actual situation, which is not limited here. So far, the key local audio segment and the non-key local audio segment in the enterprise conference audio are obtained.

[0079] In actual situations, the content of corporate meetings is coherent. In order to avoid that the local audio segments that are very relevant to the key local audio segments are treated as non-key local audio segments and are lossily compressed and transmitted to the corporate database, resulting in incomplete key information of the corporate meeting, this embodiment further analyzes the degree of coherence between each key local audio segment and its adjacent local audio segments to ensure that important and coherent corporate meeting information is accurately and completely transmitted to the corporate database. It is known that each key local audio segment must be losslessly compressed and transmitted to the corporate database. Therefore, the key local audio segments whose two adjacent local audio segments are both key local audio segments are not analyzed, and only the key local audio segments whose adjacent local audio segments are non-key local audio segments are analyzed. When the degree of coherence between a key local audio segment and its adjacent non-key local audio segment is greater, it means that the voice information between the key local audio segment and its adjacent non-key local audio segment is more relevant. Therefore, the adjacent non-key local audio segments of the key local audio segment should also be losslessly compressed and transmitted to the corporate database to ensure the integrity and accuracy of the key information of the corporate meeting.

[0080] Preferably, in some possible implementations of this embodiment, the method for obtaining the degree of coherence is as follows: for any key local audio segment, when the local audio segment adjacent to the key local audio segment is a non-key local audio segment, any non-key local audio segment adjacent to the key local audio segment is used as a reference audio segment; a speaker whose audio exists in both the key local audio segment and the reference audio segment is obtained as a reference speaker; when the audio duration of each reference speaker in the key local audio segment and the reference audio segment is longer, it indicates that the coherence between the key local audio segment and the reference audio segment is stronger. Then, according to the audio duration of each reference speaker in the key local audio segment and the reference audio segment, a second eigenvalue of each reference speaker is obtained, wherein the larger the second eigenvalue is, the longer the audio duration of the corresponding reference speaker in the key local audio segment and the reference audio segment is, which indirectly indicates that the coherence between the key local audio segment and the reference audio segment is stronger. In some possible implementations of this embodiment, the method for obtaining the second eigenvalue is: for any reference speaker, the cumulative duration of all audio durations of the reference speaker in the key local audio segment and the reference audio segment is obtained as the first duration; the cumulative result of the corresponding durations of the key local audio segment and the reference audio segment is used as the second duration; and then the ratio of the first duration to the second duration is used as the second eigenvalue of the reference speaker. At this point, the second eigenvalue of each reference speaker is obtained. The smaller the actual criticality difference between the key local audio segment and the reference audio segment, the stronger the coherence between the key local audio segment and the reference audio segment. Therefore, this embodiment obtains the coherence of the key local audio segment and the reference audio segment according to the second eigenvalue of each reference speaker and the actual criticality difference between the key local audio segment and the reference audio segment; wherein the second eigenvalue is positively correlated with the coherence, and the actual criticality difference is negatively correlated with the coherence. In order to more accurately determine the coherence between each key local audio segment and its adjacent non-key local audio segment, this embodiment specifically quantifies the coherence between each key local audio segment and its adjacent non-key local audio segment, and obtains the coherence value between each key local audio segment and its adjacent non-key local audio segment, wherein the larger the coherence value, the more relevant the voice information between the corresponding key local audio segment and its adjacent non-key local audio segment. In order to ensure that the important information of the enterprise meeting is completely stored in the enterprise database, the adjacent non-key local audio segment of the corresponding key local audio segment should be losslessly compressed and transmitted to the enterprise database. The expression of the coherence value is: ; In the formula, is the coherence value between the e-th key local audio segment and its adjacent f-th non-key local audio segment; is the number of reference speakers whose audio exists in both the e-th key local audio segment and its adjacent f-th non-key local audio segment; is the audio duration of the xth reference speaker in the eth key local audio segment; is the audio duration of the xth reference speaker in the fth non-critical local audio segment; is the duration corresponding to the e-th key local audio segment; is the duration corresponding to the fth non-critical local audio segment; For the first duration; For the second duration; is the second eigenvalue; is the actual criticality value of the e-th key local audio segment; is the actual criticality value of the fth non-critical local audio segment; is the absolute value function. It should be noted that Must be greater than 0, because Must be greater than .

[0081] It is known that the larger the coherence value is, the more the adjacent non-critical local audio segment of the corresponding critical local audio segment should be losslessly compressed and transmitted to the enterprise database. Therefore, in this embodiment, the coherence values ​​are normalized to obtain the normalized coherence value of each critical local audio segment and its adjacent non-critical local audio segment. When the normalized coherence value is greater than or equal to the preset coherence value threshold, the adjacent non-critical local audio segment of the corresponding critical local audio segment is modified to a critical local audio segment. At this point, the critical local audio segments and non-critical local audio segments in the enterprise conference audio are accurately determined.

[0082] In order to ensure the integrity and accuracy of important information in the enterprise conference audio, and at the same time improve the efficiency of transmitting the enterprise conference audio to the enterprise database and save the enterprise conference audio transmission resources, this embodiment uses the WAVPACK algorithm to losslessly compress the key local audio segments and transmit them to the enterprise database, and uses the Opus algorithm to losslessly compress the non-key local audio segments and transmit them to the enterprise database. At this point, the enterprise conference audio is transmitted to the enterprise database for storage. Among them, the WAVPACK algorithm and the Opus algorithm are well-known technologies and will not be described in detail.

[0083] In summary, this embodiment obtains the enterprise conference audio and divides it into local audio segments; obtains the criticality of the local audio segments based on the distribution of conference keywords in the local audio segments; obtains the speaker's audio importance based on the distribution of conference connecting words and conference keywords in the speaker's corresponding audio segments; corrects the criticality based on the speaker's audio duration and audio importance in the local audio segments, obtains the actual criticality, determines the critical local audio segments and non-critical local audio segments, and compresses and transmits the enterprise conference audio. The present invention determines the critical local audio segments and non-critical local audio segments, and then compresses and transmits the enterprise conference audio separately, so that during the transmission process, the accuracy of the key information of the enterprise meeting is ensured, and the efficiency of the enterprise conference audio transmission is improved.

[0084] Embodiment 2:

[0085] The present invention also proposes an enterprise data intelligent management system based on speech analysis, please refer to Figure 4 , which shows a structural diagram of an enterprise data intelligent management system based on speech analysis provided by an embodiment of the present invention. The system includes: an audio acquisition module 10, a criticality acquisition module 20, an audio importance acquisition module 30, an actual criticality acquisition module 40 and a management module 50.

[0086] The audio acquisition module 10 is used to acquire the enterprise conference audio and divide the enterprise conference audio into local audio segments of preset duration.

[0087] The criticality acquisition module 20 is used to acquire the criticality of each local audio segment according to the distribution of conference keywords in each local audio segment.

[0088] The audio importance acquisition module 30 is used to acquire the audio importance of each speaker in each enterprise meeting according to the distribution of conference conjunctions and conference keywords in the audio segment corresponding to each speaker in each enterprise meeting.

[0089] The actual criticality acquisition module 40 is used to modify the criticality of each local audio segment according to the audio duration and audio importance of each speaker in each local audio segment to acquire the actual criticality of each local audio segment.

[0090] The management module 50 is used to determine the key local audio segment and the non-key local audio segment based on the actual key degree, and perform compression transmission management on the key local audio segment and the non-key local audio segment respectively.

[0091] It should be noted that the system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the enterprise data intelligent management system based on speech analysis and the enterprise data intelligent management method based on speech analysis provided in the above embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0092] Embodiment 3:

[0093] The present invention also proposes an enterprise data intelligent management device based on speech analysis, the device includes a memory and a processor, wherein the memory stores an executable program code, and the processor is used to call and execute the executable program code to execute an enterprise data intelligent management method based on speech analysis provided in an embodiment of the present application. The device can be a chip, a component or a module, and the chip may include a connected processor and a memory; wherein the memory is used to store instructions, and when the processor calls and executes the instructions, the chip can execute an enterprise data intelligent management method based on speech analysis provided in the above embodiment.

[0094] In addition, the present application embodiment also protects a computer device, see Figure 5 The computer device includes a memory 401, a processor 402, and a computer program 403 stored in the memory 401 and running on the processor 402, wherein when the processor 402 executes the computer program 403, the computer device can execute any one of the enterprise data intelligent management methods based on voice analysis introduced above.

[0095] Embodiment 4:

[0096] This embodiment also provides a computer-readable storage medium, in which a computer program code is stored. When the computer program code is executed on a computer, the computer executes the above-mentioned related method steps to implement an enterprise data intelligent management method based on voice analysis provided in the above embodiment.

[0097] Embodiment 5:

[0098] This embodiment also provides a computer program product. When the computer program product runs on a computer, it enables the computer to execute the above-mentioned related steps to implement an enterprise data intelligent management method based on voice analysis provided in the above embodiment.

[0099] Among them, the device, computer-readable storage medium, computer program product or chip provided in this embodiment is used to execute the corresponding method provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method provided above, and will not be repeated here.

[0100] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0101] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. An enterprise data intelligent management method based on speech analysis, characterized in that: The method comprises the following steps: Obtain enterprise conference audio, and divide the enterprise conference audio into local audio segments of preset duration; According to the distribution of conference keywords in each local audio segment, the criticality of each local audio segment is obtained; According to the distribution of conference connecting words and conference keywords in the corresponding audio segments of each speaker in each enterprise meeting, the audio importance of each speaker in each enterprise meeting is obtained; According to the audio duration and audio importance of each speaker in each local audio segment, the criticality of each local audio segment is corrected to obtain the actual criticality of each local audio segment; Determine the key local audio segment and the non-key local audio segment based on the actual key degree, and perform compression transmission management on the key local audio segment and the non-key local audio segment respectively; The method for respectively compressing and transmitting the key local audio segments and the non-key local audio segments is as follows: The key local audio segment is transmitted using a lossless compression algorithm; The non-critical local audio segments are transmitted using a lossy compression algorithm.

2. The enterprise data intelligent management method based on speech analysis according to claim 1, characterized in that: The method for obtaining the criticality is: For any local audio segment, the ratio of the number of conference keywords appearing in the local audio segment to the number of all constituent words in the local audio segment is used as the keyword probability of the local audio segment; The ratio of the number of conference keyword types appearing in the local audio segment to the number of all conference keyword types is used as the keyword correlation value of the local audio segment; The criticality of the local audio segment is obtained according to the keyword probability and the keyword relevance value of the local audio segment; wherein both the keyword probability and the keyword relevance value are positively correlated with the criticality.

3. The enterprise data intelligent management method based on speech analysis as claimed in claim 2, characterized in that: The method for obtaining the audio importance is: For any speaker of any corporate meeting, the audio importance of the speaker is obtained based on the keyword relevance value of the speaker's corresponding audio segment and the number of occurrences of the conference connecting words in the speaker's corresponding audio segment; wherein, the keyword relevance value of the speaker's corresponding audio segment and the number of occurrences of the conference connecting words are both positively correlated with the audio importance.

4. The enterprise data intelligent management method based on speech analysis according to claim 1, characterized in that: The method for obtaining the actual criticality is: For any local audio segment, the proportion of the audio duration of each speaker in the local audio segment in the local audio segment is used as the first feature value of each speaker; Obtaining an important weight of each speaker in the local audio segment according to the first eigenvalue and the audio importance of each speaker in the local audio segment; wherein the first eigenvalue and the audio importance are both positively correlated with the important weight; According to the importance weight of each speaker in the local audio segment and the criticality of the local audio segment, the actual criticality of the local audio segment is obtained; wherein the importance weight and the criticality are both positively correlated with the actual criticality.

5. The enterprise data intelligent management method based on speech analysis according to claim 1, characterized in that: The method for obtaining the key local audio segment and the non-key local audio segment is: When the actual criticality meets the preset criticality range, the corresponding local audio segment is used as the critical local audio segment; When the actual criticality does not meet the preset criticality range, the corresponding local audio segment is regarded as a non-critical local audio segment.

6. The enterprise data intelligent management method based on speech analysis according to claim 1, characterized in that: The enterprise data intelligent management method based on speech analysis also includes: For any key local audio segment, when a local audio segment adjacent to the key local audio segment is a non-key local audio segment, any non-key local audio segment adjacent to the key local audio segment is used as a reference audio segment; Obtain a speaker whose audio is present in both the key local audio segment and the reference audio segment as a reference speaker; Obtaining a second feature value of each reference speaker according to the audio duration of each reference speaker in the key local audio segment and the reference audio segment; According to the second eigenvalue of each reference speaker and the actual criticality difference between the key local audio segment and the reference audio segment, the coherence degree of the key local audio segment and the reference audio segment is obtained; wherein the second eigenvalue is positively correlated with the coherence degree, and the actual criticality difference is negatively correlated with the coherence degree; When the coherence level meets a preset coherence level range, the reference audio segment is modified into a key local audio segment.

7. The enterprise data intelligent management method based on speech analysis as claimed in claim 6, characterized in that: The method for obtaining the second characteristic value is: For any reference speaker, obtain the cumulative duration of all audio durations of the reference speaker in the key local audio segment and the reference audio segment as the first duration; The cumulative result of the corresponding durations of the key local audio segment and the reference audio segment is used as the second duration; The ratio of the first duration to the second duration is used as the second characteristic value of the reference speaker.

8. The enterprise data intelligent management method based on speech analysis according to claim 1, characterized in that: The method for obtaining the conference keywords and conference connecting words is as follows: By manually setting the meeting keywords and meeting connecting words for each corporate meeting audio; among them, the meeting connecting words include summary words and emergency words.

9. The enterprise data intelligent management method based on speech analysis as claimed in claim 2, characterized in that: The Stanford CoreNLP natural speech processing tool is used to obtain the constituent words in the enterprise meeting audio.