Conference centralized management method and system and storage medium
Patent Information
- Application Number
- CN202411343900.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2025-01-03
Smart Images
Figure CN119252261A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of conference management, and in particular to a centralized conference management method, system and storage medium. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] A meeting mainly refers to a collective meeting held by an organization, institution or company for discussion and decision-making. The intelligent conference system is a form of meeting that uses technologies such as voice recognition, artificial intelligence, video conferencing and big data analysis to improve meeting efficiency and experience. In the specific implementation of the intelligent conference system, different equipment and services can be selected according to different meeting needs, thereby improving meeting efficiency and quality.
[0004] The current digital conference management solution can realize digital video capture for the meeting, which is convenient for the subsequent review and reproduction of the meeting. However, in actual situations, the content of the speeches of the speakers changes dynamically during the meeting. When the digital video capture management solution of the prior art is used to review and reproduce the subsequent meeting, there are problems such as unclear speaking objects and inaccurate identification of speech content, which affects the efficiency and quality of the actual meeting review and reproduction. Summary of the invention
[0005] In order to solve the above-mentioned technical problems, a centralized conference management method, system and storage medium are provided. This technical solution solves the problem that the digital video acquisition management solution of the above-mentioned prior art has unclear speaking objects and cannot accurately identify the content of the speech when reviewing and reproducing the subsequent meetings, which affects the efficiency and quality of the actual meeting review and reproduction.
[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:
[0007] A centralized conference management method, comprising:
[0008] At least one conference terminal is united into a conference terminal group and forms a communication link;
[0009] The conference terminal collects the characteristic voices of conference participants;
[0010] Using Fourier transform on the characteristic speech to obtain at least one basic characteristic speech, performing voiceprint recognition on the basic characteristic speech to obtain a characteristic voiceprint spectrum, and constructing a characteristic voiceprint spectrum library of conference participants;
[0011] Acquire conference speech in the environment, apply Fourier transform to the conference speech to obtain at least one basic actual speech, perform voiceprint recognition on the basic actual speech, and obtain an actual voiceprint spectrum;
[0012] Determine at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as the target voiceprint spectrum corresponding to the actual voiceprint spectrum, and determine the conference participant corresponding to the target voiceprint spectrum;
[0013] Select the conference terminal closest to the voice sender in the conference terminal group as the characteristic conference terminal;
[0014] Features Conference terminals share their own screen status with other conference terminals;
[0015] Use inverse Fourier transform on the basic actual speech corresponding to the actual voiceprint spectrum to obtain the denoised speech;
[0016] Based on the denoised speech, semantic recognition is performed to obtain the conference text, and the conference participants corresponding to the target voiceprint spectrum are annotated on the conference text to obtain the conference text data;
[0017] Use multi-conference terminal distributed storage technology to store conference text data.
[0018] Preferably, the at least one conference terminal is combined into a conference terminal group and forms a communication link, comprising the following steps:
[0019] Randomly select a conference terminal from at least one conference terminal as a first conference terminal;
[0020] Selecting two conference terminals with the shortest distance from the first conference terminal from the remaining conference terminals as second conference terminals, and pairing the first conference terminal with the two second conference terminals;
[0021] When the first conference terminal traverses at least one conference terminal, pairing is completed;
[0022] Each of the conference terminals establishes a connection with its paired conference terminal, and the two are combined into a conference terminal group, and the connection mode used for communication forms a communication link.
[0023] Preferably, the Fourier transform is as follows:
[0024]
[0025] Wherein, F(x) is the speech after Fourier transform, i is a unit imaginary number, e is a natural constant, f(t) is the speech before Fourier transform, f(t) is a time domain function, and t is time;
[0026] The inverse Fourier transform is as follows:
[0027]
[0028] Among them, G(t) is the speech before inverse Fourier transform, i is a unit imaginary number, e is a natural constant, g(x) is the speech after inverse Fourier transform, g(x) is a frequency domain function, and x is the frequency.
[0029] Preferably, the step of determining at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as the target voiceprint spectrum corresponding to the actual voiceprint spectrum specifically includes:
[0030] Calculate the pixel value difference between the actual voiceprint spectrum and the pixel point at the corresponding position in at least one characteristic voiceprint spectrum, and accumulate to obtain at least one difference judgment value;
[0031] When one of the difference judgment values is smaller than a preset value, the actual voiceprint spectrum is used as the target voiceprint spectrum.
[0032] Preferably, the step of performing semantic recognition based on denoised speech to obtain the conference text specifically includes:
[0033] According to the set frame duration, the denoised speech is divided into several denoised speech segments;
[0034] Perform keyword recognition on the denoised speech to obtain at least one similar keyword corresponding to each denoised speech segment;
[0035] Determine whether the similarity between the pronunciation data of the similar keyword and the denoised speech of the corresponding segment is greater than a first preset value. If so, use the similar keyword as the conference text of the denoised speech of the corresponding segment. If not, based on at least one similar keyword of the denoised speech of the adjacent segment, use a multi-level semantic recognition model to determine the conference text of the denoised speech of the corresponding segment.
[0036] Preferably, the method of using a multi-level semantic recognition model to determine the conference text of the denoised speech of the corresponding segment comprises the following steps:
[0037] Retrieve the denoised speech of the preceding segment and the denoised speech of the following segment of the denoised speech of the corresponding segment;
[0038] Determine whether the conference text has been determined in the previous denoised speech and the next denoised speech. If so, use the determined conference text as the semantic recognition text. If not, select one of at least one similar keyword from the previous denoised speech and / or the next denoised speech to group them as the semantic recognition text.
[0039] Calculate the semantic fitting index of each similar keyword corresponding to the denoised speech based on the semantic recognition formula;
[0040] Filter out similar keywords and semantically recognized texts with the largest semantic fitting index, and use them as conference texts of the corresponding segment of denoised speech, the previous segment of denoised speech, and the next segment of denoised speech, respectively;
[0041] The semantic recognition formula is specifically:
[0042] P=α×Y p +(U1+U2)
[0043] In the formula, P is the semantic fitting index of similar keywords, Y p is the similarity between the pronunciation data of similar keywords and the denoised speech of the corresponding segment, U1 is the semantic relevance between the similar keywords and the semantic recognition text of the previous segment of denoised speech, U2 is the semantic relevance between the similar keywords and the semantic recognition text of the next segment of denoised speech, and α is the speech fitting correction value.
[0044] Preferably, the method of storing conference text data using multi-conference terminal distributed storage technology includes the following steps:
[0045] Divide and slice the semantic recognition model and the semantic guidance model, and store the data evenly on multiple conference terminals;
[0046] Set up data replication and redundant backup strategies in multiple conference terminals, using master-slave replication or multi-master replication to replicate data to multiple conference terminals;
[0047] Use distributed transaction processing technology to achieve data consistency and synchronization among multiple conference terminals;
[0048] Adopt data sharding routing method to implement load balancing and performance optimization strategies in multiple conference terminals;
[0049] In multiple conference terminals, a disaster tolerance and fault recovery mechanism including fault detection and automatic switching is set up.
[0050] Furthermore, a conference centralized management system is proposed, which is characterized in that it is used to implement the above-mentioned conference centralized management method, including:
[0051] A link forming module, wherein the link forming module combines at least one conference terminal into a conference terminal group and forms a communication link;
[0052] A terminal identification module, wherein the terminal identification module selects a conference terminal in the conference terminal group that is closest to the voice sender as a characteristic conference terminal;
[0053] A synchronization module, the synchronization module is used to enable the feature conference terminal to share its own screen state with the remaining conference terminals;
[0054] A voice acquisition module, which is used to control the conference terminal to collect the voices of conference participants;
[0055] A voiceprint recognition module, which performs voiceprint recognition on basic characteristic speech to obtain a characteristic voiceprint spectrum, and performs voiceprint recognition on basic actual speech to obtain an actual voiceprint spectrum;
[0056] A Fourier transform module, wherein the Fourier transform module is used to implement Fourier transform;
[0057] An inverse Fourier transform module, wherein the inverse Fourier transform module is used to implement an inverse Fourier transform;
[0058] A speech recognition module, which is used to perform semantic recognition based on denoised speech to obtain conference text, and annotate the conference text with the conference participants corresponding to the target voiceprint spectrum to obtain conference text data;
[0059] A conference record storage module is used to store conference text data using a multi-conference terminal distributed storage technology.
[0060] Compared with the prior art, the present invention has the following beneficial effects:
[0061] The present invention proposes a centralized conference management solution. At the beginning of the conference, the characteristic voices of all conference participants are used, and a characteristic voiceprint spectrum library of conference participants is constructed. During the conference, conference voices are collected, and Fourier transform technology is used to accurately separate a number of basic actual voices. Voiceprint matching is performed on each basic actual voice, the speaker in the conference voice is identified, and voice recognition is performed to generate conference text. In this way, clear and definite conference records can be generated in real time, which can effectively ensure that the speaker object is clear and the speech content is accurately obtained when reviewing and reproducing the meeting in the later stage, thereby ensuring the efficiency of reviewing and reproducing the meeting and improving the level of conference management. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Figure 1 This is a flow chart of the centralized conference management method proposed by the present invention;
[0063] Figure 2 A flow chart of a method for forming a communication link by combining at least one conference terminal into a conference terminal group in the present invention;
[0064] Figure 3 A flow chart of a method for determining at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as a target voiceprint spectrum corresponding to the actual voiceprint spectrum in the present invention;
[0065] Figure 4This is a flow chart of a method for obtaining conference text by performing semantic recognition based on denoised speech in the present invention;
[0066] Figure 5 A flow chart of a method for determining conference text of denoised speech of corresponding segments by using a multi-level semantic recognition model in the present invention;
[0067] Figure 6 A flow chart of a method for storing conference text data using a multi-conference terminal distributed storage technology in the present invention;
[0068] Figure 7 It is a schematic diagram of the computer-readable storage medium structure of the present invention. DETAILED DESCRIPTION
[0069] Embodiment 1
[0070] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.
[0071] Reference Figure 1 As shown, a centralized conference management method includes:
[0072] At least one conference terminal is united into a conference terminal group and forms a communication link;
[0073] The conference terminal collects the characteristic voices of conference participants;
[0074] Using Fourier transform on the characteristic speech to obtain at least one basic characteristic speech, performing voiceprint recognition on the basic characteristic speech to obtain a characteristic voiceprint spectrum, and constructing a characteristic voiceprint spectrum library of conference participants;
[0075] Acquire conference speech in the environment, apply Fourier transform to the conference speech to obtain at least one basic actual speech, perform voiceprint recognition on the basic actual speech, and obtain an actual voiceprint spectrum;
[0076] Determine at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as the target voiceprint spectrum corresponding to the actual voiceprint spectrum, and determine the conference participant corresponding to the target voiceprint spectrum;
[0077] Select the conference terminal closest to the voice sender in the conference terminal group as the characteristic conference terminal;
[0078] Features Conference terminals share their own screen status with other conference terminals;
[0079] Use inverse Fourier transform on the basic actual speech corresponding to the actual voiceprint spectrum to obtain the denoised speech;
[0080] Based on the denoised speech, semantic recognition is performed to obtain the conference text, and the conference participants corresponding to the target voiceprint spectrum are annotated on the conference text to obtain the conference text data;
[0081] Use multi-conference terminal distributed storage technology to store conference text data.
[0082] At the beginning of the meeting, this solution uses the characteristic voices of all conference participants and builds a characteristic voiceprint spectrum library of conference participants. During the meeting, the conference voice is collected and the Fourier transform technology is used to accurately separate several basic actual voices. Voiceprint matching is performed on each basic actual voice, the speaker in the conference voice is identified, and voice recognition is performed to generate conference text. In this way, clear and unambiguous meeting records can be generated in real time.
[0083] Reference Figure 2 As shown, at least one conference terminal is united into a conference terminal group and forms a communication link, including the following steps:
[0084] Randomly select a conference terminal from at least one conference terminal as a first conference terminal;
[0085] Select two conference terminals with the shortest distance from the first conference terminal from the remaining conference terminals as second conference terminals, and pair the first conference terminal with the two second conference terminals;
[0086] When the first conference terminal traverses at least one conference terminal, pairing is completed;
[0087] Each conference terminal establishes a connection with its paired conference terminal, and the two are combined into a conference terminal group, and the connection mode used for communication forms a communication link.
[0088] The Fourier transform is as follows:
[0089]
[0090] Wherein, F(x) is the speech after Fourier transform, i is a unit imaginary number, e is a natural constant, f(t) is the speech before Fourier transform, f(t) is a time domain function, and t is time;
[0091] The inverse Fourier transform is as follows:
[0092]
[0093] Among them, G(t) is the speech before inverse Fourier transform, i is a unit imaginary number, e is a natural constant, g(x) is the speech after inverse Fourier transform, g(x) is a frequency domain function, and x is the frequency.
[0094] The goal of Fourier transform is to decompose characteristic speech into at least one basic characteristic speech, and to separate multiple human voices in the conference speech obtained in the environment, so as to obtain denoised speech using inverse Fourier transform, thereby ensuring that each denoised speech contains only the language of one speaker, thereby avoiding cross-interference of multiple speakers during speech recognition.
[0095] The goal of Fourier transform is to decompose characteristic speech into at least one basic characteristic speech, separate multiple human voices in the conference speech obtained in the environment, and use Fourier transform to convert the audio signal from the time domain to the frequency domain. Due to the different frequency characteristics of different human voices, human voices can be distinguished by analyzing the characteristics of different frequency components.
[0096] Reference Figure 3 As shown, determining at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as the target voiceprint spectrum corresponding to the actual voiceprint spectrum specifically includes:
[0097] Calculate the pixel value difference between the actual voiceprint spectrum and the pixel point at the corresponding position in at least one characteristic voiceprint spectrum, and accumulate to obtain at least one difference judgment value;
[0098] When one of the difference judgment values is smaller than the preset value, the actual voiceprint spectrum is used as the target voiceprint spectrum.
[0099] The voiceprint spectrum is an image similar to an infrared spectrum. Each person's voice is different, and the corresponding voiceprint spectrum is also different. Therefore, the voiceprint spectrum can be used as the basis for recognition to identify the speaker corresponding to the human voice in the conference speech.
[0100] Voiceprint is a sound wave spectrum that carries speech information and is displayed by an electroacoustic instrument. It represents the unique voice characteristics of each person and is specific and relatively stable. Voiceprint recognition technology is a biometric technology that identifies the identity of a speaker through sound, also known as speaker recognition. This technology obtains voice signals through audio acquisition equipment, and after preprocessing and feature extraction, it extracts spectral feature parameters that can characterize the speaker's specific organ structure or behavioral habits. Therefore, combined with Fourier transform, the speaker's voice can be identified.
[0101] Reference Figure 4 As shown, based on the denoised speech, semantic recognition is performed to obtain the conference text, which specifically includes:
[0102] According to the set frame duration, the denoised speech is divided into several denoised speech segments;
[0103] Perform keyword recognition on the denoised speech to obtain at least one similar keyword corresponding to each denoised speech segment;
[0104] Determine whether the similarity between the pronunciation data of the similar keyword and the denoised speech of the corresponding segment is greater than a first preset value. If so, use the similar keyword as the conference text of the denoised speech of the corresponding segment. If not, based on at least one similar keyword of the denoised speech of the adjacent segment, use a multi-level semantic recognition model to determine the conference text of the denoised speech of the corresponding segment.
[0105] Among them, the above keywords represent text features that have similar pronunciations to speech.
[0106] Reference Figure 5 As shown, using a multi-level semantic recognition model, determining the conference text of the denoised speech of the corresponding segment includes the following steps:
[0107] Retrieve the denoised speech of the preceding segment and the denoised speech of the following segment of the denoised speech of the corresponding segment;
[0108] Determine whether the conference text has been determined in the previous denoised speech and the next denoised speech. If so, use the determined conference text as the semantic recognition text. If not, select one of at least one similar keyword from the previous denoised speech and / or the next denoised speech to group them as the semantic recognition text.
[0109] Calculate the semantic fitting index of each similar keyword corresponding to the denoised speech based on the semantic recognition formula;
[0110] Filter out similar keywords and semantically recognized texts with the largest semantic fitting index, and use them as conference texts of the corresponding segment of denoised speech, the previous segment of denoised speech, and the next segment of denoised speech, respectively;
[0111] Among them, the semantic recognition formula is specifically:
[0112] P=α×Y p +(U1+U2)
[0113] In the formula, P is the semantic fitting index of similar keywords, Y p is the similarity between the pronunciation data of similar keywords and the denoised speech of the corresponding segment, U1 is the semantic relevance between the similar keywords and the semantic recognition text of the previous segment of denoised speech, U2 is the semantic relevance between the similar keywords and the semantic recognition text of the next segment of denoised speech, and α is the speech fitting correction value.
[0114] It is understandable that in an actual conference environment, due to differences in the speaker's voice pronunciation, multiple keywords may be matched with the language speech at the same time during actual language recognition. Based on this, this solution uses fuzzy recognition to identify several similar keywords. When the matching degree between similar keywords and the language reaches a certain threshold, the similar keywords are identified as language texts. Otherwise, the speaker's language text is identified based on the semantic association between the language before and after as the conference text. This greatly improves the recognition accuracy of the conference language, and can effectively meet the language recognition needs in a conference environment.
[0115] Reference Figure 6 As shown, using the multi-conference terminal distributed storage technology to store conference text data includes the following steps:
[0116] Divide and slice the semantic recognition model and the semantic guidance model, and store the data evenly on multiple conference terminals;
[0117] Set up data replication and redundant backup strategies in multiple conference terminals, using master-slave replication or multi-master replication to replicate data to multiple conference terminals;
[0118] Use distributed transaction processing technology to achieve data consistency and synchronization among multiple conference terminals;
[0119] Adopt data sharding routing method to implement load balancing and performance optimization strategies in multiple conference terminals;
[0120] In multiple conference terminals, a disaster tolerance and fault recovery mechanism including fault detection and automatic switching is set up.
[0121] This solution uses distributed storage of multiple conference terminals to store conference records, which can effectively improve the fault tolerance of conference record storage, and thus avoid the damage and loss of established conference records when some conference terminals are accidentally damaged.
[0122] Further, based on the same inventive concept as the above-mentioned centralized conference management method, this solution proposes a centralized conference management system, including:
[0123] A link forming module, wherein the link forming module combines at least one conference terminal into a conference terminal group and forms a communication link;
[0124] A terminal identification module, wherein the terminal identification module selects a conference terminal in the conference terminal group that is closest to the voice sender as a characteristic conference terminal;
[0125] A synchronization module, which is used to enable the feature conference terminal to share its own screen state with the remaining conference terminals;
[0126] A voice acquisition module, which is used to control the conference terminal to collect the voices of conference participants;
[0127] A voiceprint recognition module performs voiceprint recognition on basic characteristic speech to obtain a characteristic voiceprint spectrum, and performs voiceprint recognition on basic actual speech to obtain an actual voiceprint spectrum;
[0128] Fourier transform module, the Fourier transform module is used to implement Fourier transform;
[0129] An inverse Fourier transform module, which is used to implement an inverse Fourier transform;
[0130] The speech recognition module is used to perform semantic recognition based on denoised speech to obtain conference text, and annotate the conference text with the conference participants corresponding to the target voiceprint spectrum to obtain conference text data;
[0131] The conference record storage module is used to store conference text data using multi-conference terminal distributed storage technology.
[0132] The usage process of the above conference centralized management system is as follows:
[0133] Step 1: The link forming module combines at least one conference terminal into a conference terminal group and forms a communication link;
[0134] Step 2: The voice acquisition module controls the conference terminal to collect the characteristic voices of the conference participants, and at the same time calls the Fourier transform module to perform Fourier transform on the characteristic voices to obtain at least one basic characteristic voice, and then calls the voiceprint recognition module to perform voiceprint recognition on the basic characteristic voice to obtain the characteristic voiceprint spectrum, and builds the characteristic voiceprint spectrum library of the conference participants;
[0135] Step 3: The voice acquisition module acquires the conference voice in the environment, and calls the Fourier transform module to use Fourier transform on the conference voice to obtain at least one basic actual voice, and then calls the voiceprint recognition module to perform voiceprint recognition on the basic actual voice to obtain the actual voiceprint spectrum;
[0136] Step 4: The voiceprint recognition module determines at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as the target voiceprint spectrum corresponding to the actual voiceprint spectrum, and determines the conference participant corresponding to the target voiceprint spectrum;
[0137] Step 5: The terminal identification module selects the conference terminal closest to the voice sender in the conference terminal group as the characteristic conference terminal, and calls the synchronization module to make the characteristic conference terminal share its screen state with the remaining conference terminals;
[0138] Step 6: The inverse Fourier transform module uses the inverse Fourier transform on the basic actual speech corresponding to the actual voiceprint spectrum to obtain the denoised speech;
[0139] Step 7: The speech recognition module performs semantic recognition based on the denoised speech to obtain the conference text, and annotates the conference text with the conference participants corresponding to the target voiceprint spectrum to obtain the conference text data;
[0140] Step 8: The conference record storage module uses multi-conference terminal distributed storage technology to store conference text data.
[0141] Figure 7 Schematic diagram of a computer-readable storage medium structure provided by an embodiment of the present application. Figure 7 As shown, a computer-readable storage medium 600 according to an embodiment of the present application is shown. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by the processor, a conference centralized management method according to an embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.
[0142] In summary, the advantages of the present invention are: effectively ensuring that when reviewing and reproducing the subsequent meeting, the speaking object is clearly identified and the speech content is accurately obtained, thereby ensuring the efficiency of the meeting review and reproduction and improving the level of meeting management.
[0143] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.
Claims
1. A centralized conference management method, characterized in that: include: At least one conference terminal is united into a conference terminal group and forms a communication link; The conference terminal collects the characteristic voices of conference participants; Using Fourier transform on the characteristic speech to obtain at least one basic characteristic speech, performing voiceprint recognition on the basic characteristic speech to obtain a characteristic voiceprint spectrum, and constructing a characteristic voiceprint spectrum library of conference participants; Acquire conference speech in the environment, apply Fourier transform to the conference speech to obtain at least one basic actual speech, perform voiceprint recognition on the basic actual speech, and obtain an actual voiceprint spectrum; Determine at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as the target voiceprint spectrum corresponding to the actual voiceprint spectrum, and determine the conference participant corresponding to the target voiceprint spectrum; Select the conference terminal closest to the voice sender in the conference terminal group as the characteristic conference terminal; Features Conference terminals share their own screen status with other conference terminals; Use inverse Fourier transform on the basic actual speech corresponding to the actual voiceprint spectrum to obtain the denoised speech; Based on the denoised speech, semantic recognition is performed to obtain the conference text, and the conference participants corresponding to the target voiceprint spectrum are annotated on the conference text to obtain the conference text data; Use multi-conference terminal distributed storage technology to store conference text data.
2. A centralized conference management method according to claim 1, characterized in that: The at least one conference terminal is combined into a conference terminal group and forms a communication link, including the following steps: Randomly select a conference terminal from at least one conference terminal as a first conference terminal; Selecting two conference terminals with the shortest distance from the first conference terminal from the remaining conference terminals as second conference terminals, and pairing the first conference terminal with the two second conference terminals; When the first conference terminal traverses at least one conference terminal, pairing is completed; Each of the conference terminals establishes a connection with its paired conference terminal, and the two are combined into a conference terminal group, and the connection mode used for communication forms a communication link.
3. A centralized conference management method according to claim 2, characterized in that: The Fourier transform is as follows: Wherein, F(x) is the speech after Fourier transform, i is a unit imaginary number, e is a natural constant, f(t) is the speech before Fourier transform, f(t) is a time domain function, and t is time; The inverse Fourier transform is as follows: Among them, G(t) is the speech before inverse Fourier transform, i is a unit imaginary number, e is a natural constant, g(x) is the speech after inverse Fourier transform, g(x) is a frequency domain function, and x is the frequency.
4. A centralized conference management method according to claim 3, characterized in that: The step of determining at least one characteristic voiceprint spectrum whose difference from the actual voiceprint spectrum is less than a preset value as the target voiceprint spectrum corresponding to the actual voiceprint spectrum specifically includes: Calculate the pixel value difference between the actual voiceprint spectrum and the pixel point at the corresponding position in at least one characteristic voiceprint spectrum, and accumulate to obtain at least one difference judgment value; When one of the difference judgment values is smaller than a preset value, the actual voiceprint spectrum is used as the target voiceprint spectrum.
5. A centralized conference management method according to claim 4, characterized in that: The method of performing semantic recognition based on denoised speech to obtain the conference text specifically includes: According to the set frame duration, the denoised speech is divided into several denoised speech segments; Perform keyword recognition on the denoised speech to obtain at least one similar keyword corresponding to each denoised speech segment; Determine whether the similarity between the pronunciation data of the similar keyword and the denoised speech of the corresponding segment is greater than a first preset value. If so, use the similar keyword as the conference text of the denoised speech of the corresponding segment. If not, based on at least one similar keyword of the denoised speech of the adjacent segment, use a multi-level semantic recognition model to determine the conference text of the denoised speech of the corresponding segment.
6. A centralized conference management method according to claim 5, characterized in that: The method of using a multi-level semantic recognition model to determine the conference text of the denoised speech of the corresponding segment comprises the following steps: Retrieve the denoised speech of the preceding segment and the denoised speech of the following segment of the denoised speech of the corresponding segment; Determine whether the conference text has been determined in the previous denoised speech and the next denoised speech. If so, use the determined conference text as the semantic recognition text. If not, select one of at least one similar keyword from the previous denoised speech and / or the next denoised speech to group them as the semantic recognition text. Calculate the semantic fitting index of each similar keyword corresponding to the denoised speech based on the semantic recognition formula; Filter out similar keywords and semantically recognized texts with the largest semantic fitting index, and use them as conference texts of the corresponding segment of denoised speech, the previous segment of denoised speech, and the next segment of denoised speech, respectively; The semantic recognition formula is specifically: P=α×Y p +(U1+U2) In the formula, P is the semantic fitting index of similar keywords, Y p is the similarity between the pronunciation data of similar keywords and the denoised speech of the corresponding segment, U1 is the semantic relevance between the similar keywords and the semantic recognition text of the previous segment of denoised speech, U2 is the semantic relevance between the similar keywords and the semantic recognition text of the next segment of denoised speech, and α is the speech fitting correction value.
7. A centralized conference management method according to claim 6, characterized in that: The method of storing conference text data using multi-conference terminal distributed storage technology includes the following steps: Divide and slice the semantic recognition model and the semantic guidance model, and store the data evenly on multiple conference terminals; Set up data replication and redundant backup strategies in multiple conference terminals, using master-slave replication or multi-master replication to replicate data to multiple conference terminals; Use distributed transaction processing technology to achieve data consistency and synchronization among multiple conference terminals; Adopt data sharding routing method to implement load balancing and performance optimization strategies in multiple conference terminals; In multiple conference terminals, a disaster tolerance and fault recovery mechanism including fault detection and automatic switching is set up.
8. A centralized conference management system, characterized in that: The method for implementing the centralized conference management method according to any one of claims 1 to 7 comprises: A link forming module, wherein the link forming module combines at least one conference terminal into a conference terminal group and forms a communication link; A terminal identification module, wherein the terminal identification module selects a conference terminal in the conference terminal group that is closest to the voice sender as a characteristic conference terminal; A synchronization module, the synchronization module is used to enable the feature conference terminal to share its own screen state with the remaining conference terminals; A voice acquisition module, which is used to control the conference terminal to collect the voices of conference participants; A voiceprint recognition module, which performs voiceprint recognition on basic characteristic speech to obtain a characteristic voiceprint spectrum, and performs voiceprint recognition on basic actual speech to obtain an actual voiceprint spectrum; A Fourier transform module, wherein the Fourier transform module is used to implement Fourier transform; An inverse Fourier transform module, wherein the inverse Fourier transform module is used to implement an inverse Fourier transform; A speech recognition module, which is used to perform semantic recognition based on denoised speech to obtain conference text, and annotate the conference text with the conference participants corresponding to the target voiceprint spectrum to obtain conference text data; A conference record storage module is used to store conference text data using a multi-conference terminal distributed storage technology.
9. A computer-readable storage medium storing a computer-readable program, characterized in that: When the computer-readable program is executed by a processor, the method for storing video files in a relational database based on version sequences according to any one of claims 1 to 7 is implemented.