A method and system for automatically generating a conference minutes driven by an intelligent conference terminal

Abstract

The application relates to the technical field of intelligent conference terminals and discloses a kind of intelligent conference terminal driven meeting minutes automatic generation method and system;Method includes: obtaining the identity list of attending, collecting sound data and extracting voiceprint features;Calculate identity features and generate a unique list serial number;Convert sound data into a text conference record and associate identity information;Using continuous Chinese encoding algorithm to convert the conference record into actual encoding;Based on the serial number of the list, the encoding is dynamically shifted and encrypted, and an encrypted record is generated;Store the identity list, serial number, encrypted record and corresponding relationship to form a meeting minutes. The system supports dynamic coding adjustment of newly added participants during the meeting, and optimizes the encryption effect through weighted voiceprint feature calculation, data volume proportion anti-correlation adjustment offset value and other methods. When decrypting, the original conference record is restored through the reverse algorithm, ensuring the safety of the content while realizing the automatic generation and efficient sharing of the minutes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent conference terminals, in particular to a conference summary automatic generation method and system driven by an intelligent conference terminal. BACKGROUND

[0002] With the acceleration of enterprise digitization process and the increase of cross-regional cooperation, the demand for real-time processing, accurate retention and rapid sharing of conference content is urgent, and intelligent transformation can realize automatic generation of conference summary to support rapid landing of decisions.

[0003] Current intelligent transformation of conference scenarios has gradually applied technologies such as intelligent conference terminals, speech recognition, and natural language processing. Terminals can realize real-time speech transcription and distinguish between multiple speakers, and some systems can automatically extract conference highlights and generate preliminary summary frameworks. At the same time, cloud platforms support conference data storage and cross-end sharing, and AI algorithms optimize to improve the accuracy of summary generation.

[0004] Existing conference intelligent solutions mainly focus on data transmission and storage links, and do not modify the conference content itself. Sensitive information exists in its original form, and once the terminal or platform is compromised, information is easily leaked, and encryption effect is limited. SUMMARY

[0005] In order to improve the encryption effect of conference content, the present application provides a conference summary automatic generation method and system driven by an intelligent conference terminal.

[0006] In a first aspect, the present application provides a conference summary automatic generation method driven by an intelligent conference terminal, which adopts the following technical solution:

[0007] A conference summary automatic generation method driven by an intelligent conference terminal, comprising the following steps:

[0008] Obtain an identity list of participants in a conference;

[0009] Real-time collection of conference voice data, identification of voiceprint information corresponding to the identity list from the voice data, the voiceprint information including fundamental frequency information and resonance information;

[0010] Calculate identity features according to the fundamental frequency information and the resonance information, and sort the identity features to obtain a list number corresponding to the identity list, the list number being non-repetitive;

[0011] Identify the voice data into conference record data with text content, and associate the identity list with the conference record data;

[0012] Convert the text content in the conference record data into actual encoding according to a Chinese encoding algorithm with continuous encoding range.

[0013] According to the association relationship, the actual encoding of the conference record data is shifted to obtain an encrypted encoding according to the list number corresponding to the identity list;

[0014] According to the encrypted encoding, a corresponding encrypted record is matched from the Chinese encoding algorithm;

[0015] According to the association relationship, a content corresponding relationship between the list number and the encrypted record is obtained;

[0016] The identity list, the list number, the encrypted record and the content corresponding relationship are stored as conference minutes data.

[0017] By adopting the above technical solution, the identity list of participating in the conference is obtained, the voiceprint information containing the fundamental frequency information and the resonance information is identified combined with the real-time collected conference sound data, the identity feature is calculated and sorted to obtain a unique list number, and the identity association basis for encryption is laid; after the conference record data identified by the sound data is associated with the identity list, the actual encoding is converted from the Chinese encoding algorithm with continuous encoding range, and then the actual encoding is shifted based on the list number to obtain the encrypted encoding, the encrypted record is matched, and the content corresponding relationship between the list number and the encrypted record is established, and finally the identity list, the list number, the encrypted record and the content corresponding relationship are stored as conference minutes data. The whole process makes the conference content exist in the form of encrypted encoding, rather than the original text, and can be decrypted only by corresponding identity and number; if the data is obtained by others, each character in the text content converted is a character after offset, and finally the obtained content is also wrong. The sensitive information is prevented from being leaked when the terminal or platform is attacked, and the encryption effect of the conference content is greatly improved.

[0018] Optionally, the method further comprises the following steps:

[0019] The conference minutes data is obtained;

[0020] The identity list, the list number, the encrypted record and the content corresponding relationship are extracted from the conference minutes data;

[0021] According to the encrypted record and the inverse algorithm of the Chinese encoding algorithm, the encrypted encoding is matched;

[0022] According to the encrypted record and the content corresponding relationship, the list number and the encrypted encoding are matched;

[0023] According to the corresponding list number, the encrypted encoding is reversely shifted to obtain the actual encoding;

[0024] According to the actual encoding and the inverse algorithm of the Chinese encoding algorithm, the conference record data is matched.

[0025] convert the content correspondence into the association relationship;

[0026] output the identity list, the association relationship and the meeting record data.

[0027] By adopting the technical solution, identity list, list number and other elements are extracted, the encrypted record is restored to encrypted code by using the reverse algorithm of Chinese coding algorithm, and the actual code is obtained by reverse shifting of the list number, and finally the original meeting record data is matched and the association relationship is converted. This process strictly follows the reverse operation of the forward encryption path, ensures the accuracy and reliability of the decryption process, completely restores the original meeting content and identity association, and realizes the safe restoration of the meeting minutes data.

[0028] Optionally, the method further comprises the following steps:

[0029] If the identity characteristic values of any two corresponding identity lists are the same, the values of the fundamental frequency information corresponding to the two identity characteristics are compared, and the identity characteristic with a larger value of the fundamental frequency information is increased;

[0030] If the identity characteristic values of any two corresponding identity lists are the same, and the values of the corresponding fundamental frequency information are also the same, the values of the resonance information corresponding to the two characteristics are compared, and the identity characteristic with a larger value of the resonance information is increased.

[0031] By adopting the technical solution, when the identity characteristic values are the same, the fundamental frequency information values are compared first, and the identity characteristic value of the one with a larger fundamental frequency value is increased; if the fundamental frequency is still the same, the resonance information values are compared and the one with a larger resonance value is increased. The differential processing of the identity characteristic values ensures the uniqueness of identity recognition, so that the identity characteristic corresponds to the meeting participant one by one, which is beneficial to ensure the uniqueness and accuracy of the list number.

[0032] Optionally, the method further comprises the following steps:

[0033] According to the fundamental frequency information and the resonance information, the identity characteristic is calculated by using the weighted average method;

[0034] According to the positive correlation adjustment of the number of identity lists, the weight value of the fundamental frequency information is adjusted; the smaller the number of identity lists, the smaller the weight value of the fundamental frequency information; the larger the number of identity lists, the larger the weight value of the fundamental frequency information;

[0035] According to the positive correlation adjustment of the total data amount of the sound data, the weight value of the resonance information is adjusted; the smaller the total data amount of the sound data, the smaller the weight value of the resonance information; the larger the total data amount of the sound data, the larger the weight value of the resonance information.

[0036] By adopting the technical scheme, when the number of participants increases, the fundamental frequency information weight is increased, and the individual feature distinction degree is improved; the total data amount of the sound data is increased, the resonance information weight is increased, and the feature stability in a complex environment is enhanced, and the two weights can be adaptively adjusted according to the actual situation of the meeting.

[0037] Optionally, the method further comprises the following steps:

[0038] According to the data amount proportion of the text content corresponding to the identity list in the meeting record data, the shift offset value of the identity list is inversely related to the adjustment; the smaller the data amount proportion is, the greater the shift offset value of the identity list is; the greater the data amount proportion is, the smaller the shift offset value of the identity list is.

[0039] By adopting the technical scheme, shift = list number + shift offset value; the shift offset value is inversely related to the adjustment according to the data amount proportion of the content corresponding to the identity; the more the speech is, the less the encrypted offset is, dynamic differential encryption is realized, the probability and amount of offset out of range are reduced, and the overall accuracy of the meeting minutes is improved.

[0040] Optionally, the method further comprises the following steps:

[0041] If the identity list is added in the process of generating the meeting record data, the time sorting is performed;

[0042] In the first period of time, one meeting minutes data corresponding to the current M identity lists is generated;

[0043] In the second period of time, another meeting minutes data corresponding to the newly added N identity lists is generated, wherein the sorting position corresponding to the newly added list number is M.

[0044] By adopting the technical scheme, the time sorting mechanism realizes dynamic personnel management: when a participant is added, the meeting minutes data of the current M identity lists is completed first, and then the newly added N identity is processed separately, and the sorting position is automatically connected to the original sequence; the interference of dynamic adjustment of participants on the encryption process is reduced, the data of new and old participants is independently encrypted and the serial number is continuous, the system expansibility and the flexibility of meeting content encryption are improved.

[0045] Optionally, the method further comprises the following steps:

[0046] If N identity lists are added in the process of generating the meeting record data, the time sorting is performed;

[0047] In the first period of time, the list number of the current M identity lists is generated, wherein the sorting position of the M identity lists is enlarged by N+1 times;

[0048] generating the list serial numbers of the newly added N identity lists in a second time period;

[0049] inserting the current M identity lists into the newly added N identity lists, reordering according to the corresponding list serial numbers, and then generating the conference minutes data.

[0050] By adopting the above technical solution, the identity list corresponding to the newly added participant is processed through time-sharing sorting and serial number reservation mechanism; the current M list serial numbers are enlarged by N times to reserve space, then the newly added N serial numbers are generated, and finally the reordering is inserted; this is conducive to reducing the interference of dynamic adjustment of participants on the encryption process, ensuring that the data of new and old participants is independently encrypted and the serial numbers are continuous, and improving the system expansibility and the flexibility of conference content encryption.

[0051] Optionally, the method further comprises the following steps:

[0052] obtaining the parity attribute according to the total amount of the list serial numbers;

[0053] If the parity attribute is odd, the shift is offset along the first direction, otherwise, the shift is offset along the second direction.

[0054] By adopting the above technical solution, the shift direction is distinguished by the parity attribute of the total amount of list serial numbers, odd for the first direction and even for the second direction. This increases the randomness and complexity of the encryption coding shift, avoids the regularity of a single direction, improves the anti-cracking ability of conference content encryption, and enhances the information security.

[0055] In a second aspect, the present application provides a conference minutes automatic generation system driven by an intelligent conference terminal, which adopts the following technical solution:

[0056] A conference minutes automatic generation system driven by an intelligent conference terminal, comprising a processor, wherein the processor executes the steps of the conference minutes automatic generation method driven by the intelligent conference terminal according to any one of the above.

[0057] In summary, the present application includes at least one of the following beneficial technical effects: improving conference content encryption effect, generating a unique list serial number through voiceprint recognition, combining Chinese code shift encryption, making conference content exist in encrypted form, avoiding leakage of original sensitive information. Realize the safe and efficient automatic generation and restoration of meeting minutes, the forward encryption and reverse decryption process correspond, ensure the accuracy of decryption, restore the conference content and identity association. Optimize identity feature recognition, solve feature conflict and dynamically adjust weight to ensure the uniqueness of identity features and improve the accuracy of encrypted basic data. Dynamically adjust encryption strategy, such as adjust offset value according to data volume ratio, shift direction according to total number of serial numbers, enhance encryption complexity and anti-cracking ability. Support dynamic management of meeting participants, two time-sharing sorting mechanisms deal with new personnel, ensure the stability of encryption process, and improve system scalability and flexibility. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 is a step diagram of a conference minutes automatic generation method driven by an intelligent conference terminal.

[0059] Figure 2 is a step diagram of decrypting the encrypted conference minutes data to extract and output the original identity list, the association between identity and meeting records, and the meeting record data.

[0060] Figure 3 is a step diagram when the identity feature values are the same. DETAILED DESCRIPTION

[0061] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0062] In the description of the present specification, the description of the terms "some embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the described embodiments or examples are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0063] The present application embodiment discloses a conference minutes automatic generation method driven by an intelligent conference terminal, referring to Figure 1 , including the following steps:

[0064] Identity information of participants is manually entered through the interactive interface of the intelligent conference terminal, including name, position, etc., or automatically synchronized through the docking function of the terminal with external platforms such as enterprise OA systems and party building systems, such as importing participant information of government meetings from the "Three Big Systems". The identity list is stored in the local encrypted partition of the terminal and synchronized to the cloud meeting service platform as the basis for subsequent voiceprint matching.

[0065] The intelligent conference terminal collects conference sound data in real time through an omnidirectional microphone array (with a pickup range of 15 meters, supporting clear collection of multiple speakers in a closed space). After the collected audio signals are preprocessed by the terminal processing module for noise reduction and echo cancellation, the voiceprint information corresponding to the identity list is extracted by calling the speech small model based on the light big model training and the voiceprint recognition engine:

[0066] Fundamental frequency information: the fundamental frequency of the speaker's voice is extracted through speech spectrum analysis, reflecting the characteristics of vocal cord vibration, such as male fundamental frequency of about 85-180Hz and female fundamental frequency of about 165-255Hz;

[0067] Resonance information: analysis of the resonance peak parameters in the sound spectrum, such as the first resonance peak F1 and the second resonance peak F2, reflecting the characteristics of the vocal tract structure.

[0068] The terminal compares the extracted voiceprint information with the pre-stored voiceprint template in the identity list, establishing a correspondence between the voiceprint and the identity; the voiceprint template is a template temporarily generated for the first participant.

[0069] Based on the extracted fundamental frequency information F0 and resonance information (F1, F2), the identity characteristic value is calculated, and a weighted algorithm is used to calculate the identity characteristic value: identity characteristic value = ω1 × F0 + ω2 × (F1 + F2) / 2; where the weights ω1 and ω2 are dynamically adjusted by the light big model according to the conference scene, such as increasing ω1 to strengthen individual differences in multi-person meetings. The identity characteristic values of all participants are sorted in ascending order to generate a non-repeating list number, such as 1, 2, 3... N, which is mapped one-to-one with the identity list to ensure the uniqueness of the subsequent encrypted association.

[0070] The terminal transmits the collected sound data to the speech transcription service in real time, calls a high-precision speech recognition model and a pre-set industry professional vocabulary, such as "futures contract" in financial meetings and "drug ingredients" in medical meetings, and converts the speech into text form of conference record data. At the same time, through the voiceprint recognition result, each piece of text record is associated with the corresponding identity list, such as marking "[serial number 3]: discuss annual budget adjustment plan", and the association relationship is stored in the terminal memory.

[0071] Adopt a Chinese encoding algorithm with continuous encoding range, such as GB2312 encoding, which covers 6763 commonly used Chinese characters, and convert each character in the meeting record data into the corresponding binary actual encoding. For example, the GB2312 encoding corresponding to the character "会" is 0xA1A2, which is converted into the actual encoding "1010000110100010"; professional terms (such as "三重一大") are preferentially matched with the encoding in the professional vocabulary built into the terminal to ensure the accuracy of term conversion.

[0072] Based on the previous association relationship, the terminal traverses the meeting record data and performs a shift operation on the actual encoding of each character according to the following rules: Encrypted encoding = Actual encoding + List serial number × Offset coefficient; among which, the offset coefficient can be dynamically set according to the meeting confidentiality level. For example, the offset coefficient for a top-secret meeting is 3, and for an ordinary meeting is 1; the shift direction is determined by the terminal communication module in combination with the transmission protocol of the APN encryption channel, such as determining the direction based on the key characteristics of the SM2 encryption algorithm. For example, the character "议" corresponding to the speaker with serial number 2 has an actual encoding of 0xA2B3, and after shifting, the encrypted encoding 0xA2B5 is generated.

[0073] Reverse query the Chinese encoding algorithm table according to the encrypted encoding to obtain the corresponding character content (i.e., the encrypted record). For example, the encrypted encoding 0xA2B5 corresponds to the character "屹", which is different from the original character "议". At the same time, the terminal establishes a mapping between the list serial number and the encrypted record based on the association relationship between the identity and the record, such as "Serial number 2 → Encrypted paragraph P2", to ensure traceability to the original speaker during decryption.

[0074] The terminal packs the identity list, list serial number, encrypted record, and content correspondence relationship into a meeting minutes data packet, generates a data check value using the SM3 encryption algorithm, and uploads it to the cloud storage through an exclusive channel isolated by the air gap technology. The cloud storage is encrypted using the SM2 algorithm. The terminal supports the "burn after reading" mode. Users can choose to automatically destroy the original voice and encoding data cached locally after the meeting, and only retain the encrypted meeting minutes data packet to further improve data security.

[0075] By integrating the hardware capabilities of the intelligent meeting terminal (such as omnidirectional sound pickup, voiceprint collection) with the algorithm advantages of the Guangming large model, such as professional vocabulary matching, dynamic encryption strategy, realize the full-process encrypted generation of meeting minutes: the meeting content exists in the form of encrypted encoding. Unauthorized access can only obtain scrambled characters, such as "议" becoming "屹". It is necessary to combine the identity list and the list serial number to decrypt through reverse shifting, solving the risk of leakage of the original content; relying on the professional vocabulary and high-precision speech recognition to ensure the industry adaptability and accuracy of the meeting record, providing a reliable basis for the restoration of the encrypted content; compatibility with enterprise OA, party building systems, etc., supporting the synchronization of identity lists and the cross-platform transfer of meeting minutes, meeting the efficient collaboration needs of multi-scenario meetings.

[0076] Referring to Figure 2 , the method further comprises the following steps:

[0077] The cached data can be called by the local storage module of the intelligent conference terminal, or downloaded from the cloud conference service platform through the APN encryption channel. The SM2 encryption algorithm is used to verify the terminal authority during the downloading process, so as to ensure that only authorized equipment can be accessed. After the identity list, list number, encryption record and content corresponding relationship are extracted from the conference summary data, the natural language processing capability of the Guangming big model is relied on to analyze the data structure.

[0078] When the encryption record and the reverse algorithm of the Chinese encoding algorithm are matched, the system will call the preset professional word library for auxiliary verification; for example, in the financial conference, the "futures contract" appearing in the "encryption record" will preferentially match the reverse encoding rule of "futures contract" in the professional word library, correct the encoding matching deviation, and improve the accuracy of encryption code restoration.

[0079] In the corresponding link of list number and encryption code according to the encryption record and the content corresponding relationship, the historical association data (participant voiceprint and serial number mapping log cached locally by the terminal) of voiceprint recognition is combined for secondary verification to avoid the mismatch of serial number and code caused by data transmission error.

[0080] When the encryption code is reversely shifted according to the corresponding list number, the reverse operation of the shift rule of forward encryption is strictly followed: if the forward encryption is "actual code + list number × offset coefficient", the reverse shift is "encryption code - list number × offset coefficient", and the shift direction remains the same as the forward direction, such as the first direction offset in the forward direction, and the reverse direction also moves back in the first direction. In this process, the terminal processing module will call a high-performance processor for parallel computing to ensure efficient completion of the reverse shift of a large number of encryption codes.

[0081] When the conference record data is matched according to the actual code and the reverse algorithm of the Chinese encoding algorithm, the professional word library is also linked for semantic correction, for example, the "conference" corresponding to the actual code is corrected to "meeting", to ensure that the restored text content conforms to the industry expression habit.

[0082] After the content corresponding relationship is converted into the association relationship, the system will refer to the preset template in the conference service, such as the formal conference summary template, to automatically complete the association format of the identity list and the conference record data, such as marking "[name 3 (serial number 3)]: speech content", and finally output through the terminal interaction interface or export in the format of Word, PDF, etc. The exported file is encrypted by SM3 algorithm to prevent tampering during transmission. If the conference starts the "burn after listening" mode, the terminal will automatically clear the temporary data generated during the decryption process after the output is completed, and only the original encrypted conference summary data is retained.

[0083] This decryption process fully utilizes the hardware computing power (such as high-performance processors) and software capabilities (such as professional dictionaries and large model algorithms) of the intelligent conference terminal. Through strict symmetry with the forward encryption step, it realizes the accurate restoration of the meeting minutes data, ensuring information security and the professionalism and integrity of the decrypted content, meeting the accuracy and efficiency requirements of different industries for conference data decryption.

[0084] In actual conference scenarios, there may be special cases where the identity characteristic values are the same. This method sets up a rigorous processing mechanism for this situation, and the method further includes the following steps:

[0085] When the system calculates the identity characteristic values, if it finds that the identity characteristics of any two identity lists have the same value, the intelligent conference terminal will immediately start the identity characteristic differentiation process. The terminal will first compare the values of the fundamental frequency information corresponding to the two identity characteristics. Taking the high-precision voice analysis module built into the intelligent conference terminal as an example, this module can accurately measure the fundamental frequency of the sound, with an accuracy of ±0.1 Hz. Assuming that the identity characteristic values of participant A and participant B are the same, when comparing the fundamental frequency information, if it is found that the fundamental frequency information value of participant A is 120 Hz and the fundamental frequency information value of participant B is 125 Hz, the system will increase the identity characteristic value of the participant B whose fundamental frequency information value is larger, and the specific increase can be determined according to the system preset adjustment factor, such as setting the adjustment factor to 0.05, then the identity characteristic value of participant B will be increased by 0.05 times based on the original value, thereby realizing the differentiation of the identity characteristic values of the two.

[0086] If after comparing the fundamental frequency information, it is found that the identity characteristics of any two identity lists still have the same value, and the corresponding fundamental frequency information values are also completely identical, the system will further compare the values of the resonance information corresponding to the two characteristics. The intelligent conference terminal can accurately analyze the resonance peak information of the sound through advanced acoustic algorithms, and the measurement error of the first resonance peak (F1) and the second resonance peak (F2) can be controlled within ±5 Hz. For example, when the fundamental frequencies of two participants are the same, and the resonance information is continuously compared, if the first resonance peak F1 value of participant C is 800 Hz and the second resonance peak F2 value is 1800 Hz, and the first resonance peak F1 value of participant D is 820 Hz and the second resonance peak F2 value is 1850 Hz, the system will increase the identity characteristic value of the participant D whose resonance information value is larger, and the increase is also performed according to the adjustment factor, to ensure that the identity characteristic values differ.

[0087] By comparing the fundamental frequency information value first, the identity feature value of the one with larger fundamental frequency value is promoted; if the fundamental frequency is still the same, the resonance information value is compared and the identity feature value of the one with larger resonance value is promoted, which can effectively ensure the uniqueness of identity recognition. Only when the identity features of each participant are completely different from those of others, can the identity features be one-to-one corresponding to the participants, thereby providing a solid foundation for subsequent generation of list serial number, effectively ensuring the uniqueness and accuracy of the list serial number, and providing reliable protection for the smooth development of the entire conference summary automatic generation and encryption process.

[0088] Referring to Figure 3 , the method further comprises the following steps:

[0089] In calculating the identity feature, the weighted average formula identity feature = ω1 x fundamental frequency information + ω2 x resonance information is used for calculation with the support of the high-performance processor built in the intelligent conference terminal and the algorithm of the Guangming Big Model; wherein ω1 is the weight value of the fundamental frequency information, ω2 is the weight value of the resonance information, and ω1 + ω2 = 1.

[0090] The weight adjustment mechanism is deeply adapted to the actual application scene of the product: when the number of the identity list changes, the system will judge the conference scale based on the terminal, such as identifying whether it is a small department meeting or a large industry summit from the number of the identity list, and positively adjusting the weight value of the fundamental frequency information. For example, in a small enterprise internal meeting with only 3 people, the fundamental frequency information weight value ω1 may be set to 0.3, at which time more reliance is placed on the resonance information to distinguish individuals; while in a large government agency work conference with 20 people, ω1 will be raised to 0.7, by strengthening the fundamental frequency, a prominent individual voice feature, the difference between male and female fundamental frequencies is obvious, which improves the voiceprint recognition accuracy in a multi-person scene, and is consistent with the functional characteristics of the product of distinguishing speakers.

[0091] For the weight adjustment of the resonance information, the system will dynamically adjust according to the total data amount of the sound data; the total data amount is calculated from the audio duration and sampling rate collected by the terminal microphone array. When the conference time is short and the speech is sparse, resulting in a small amount of sound data, such as a 10-minute quick meeting, the resonance information weight value ω2 may be as low as 0.2, reducing the interference of environmental noise on feature extraction; while in a 2-hour financial institution investment decision-making meeting, with the increase of the sound data amount, a large amount of continuous speech containing discussions and debates, ω2 will be raised to 0.6, taking advantage of the stability of the resonance information reflecting the stable characteristics of the vocal tract structure in long-time speech, to enhance the anti-interference ability of the voiceprint features in complex environments, and cooperating with the high-sensitivity microphone array of the product, to further ensure the accuracy of identity recognition.

[0092] This weight adaptive adjustment mechanism fully combines the hardware performance of the product (such as the pickup ability of the microphone array and the real-time computing ability of the processor) and the algorithm advantage (the intelligent judgment of the Guangming large model on the conference scene), so that the calculation of the identity feature can adapt to the needs of conferences of different scales and cope with different data volume speech environments, providing reliable basic feature data for the generation of the list number and the encryption of the conference content, and effectively supporting the core goal of the product in efficient, accurate and secure conference experience.

[0093] The method further comprises the following steps:

[0094] After generating the conference record data, the intelligent conference terminal will use the built-in data analysis module to calculate the proportion of the text content of each participant (corresponding to the identity list) in the total conference record data, such as the speaking content of participant A accounting for 30%, and participant B accounting for 5%. Based on this proportion, the terminal calls the dynamic encryption strategy module of the Guangming large model, adjusts the shift offset value according to the anti-correlation rule, and forms the encryption formula "shift = list number + shift offset value".

[0095] For example, at an investment decision-making meeting of a financial institution, if an analyst (identity list number 5) speaking content accounts for 40% (large data volume), the shift offset value may be adjusted to 1, and the final shift is 5+1=6; while a listener (number 8) speaking accounts for only 5% (small data volume), the offset value may be set to 5, and the shift is 8+5=13. This adjustment mechanism is linked with the professional vocabulary function of the product: for the industry term-intensive content with a large proportion, such as "futures contract" and "risk exposure", small offset values are used to reduce coding conversion errors and ensure the accuracy of professional vocabulary after decryption; for the scattered speech with a small proportion, large offset values are used to enhance the encryption strength, which meets the product advantage of "high security".

[0096] At the same time, the high-performance processor of the terminal will real-time check whether the encrypted code after the shift is within the effective range of the Chinese coding algorithm, such as the coding interval of GB2312. The anti-correlation adjustment mechanism can reduce the probability of coding overflow caused by large offset: the core content with more speaking has small offset, reducing the risk of out-of-range; the content with less speaking, even with large offset, has small data volume, and the total amount of out-of-range is controllable. This design not only adapts to the performance requirements of the product "supporting 20 conference rooms concurrent processing", but also improves the overall security and accuracy of the conference minutes through dynamic differentiated encryption, and together with functions such as "listen and burn" and "exclusive security encryption", it builds a multi-level data protection system.

[0097] The method further comprises the following steps:

[0098] When temporary participants appear during a meeting, such as the department head joining an enterprise meeting halfway or the additional attendees in a government meeting, the intelligent conference terminal obtains the new identity list through real-time synchronization with external platforms such as enterprise OA systems and party building systems, and triggers the time-sharing sorting mechanism. The terminal's processing module first locks the meeting data for the first period of time, and based on the current M identity list, such as the original 10 participants, calls the conference minutes generation engine of the Guangming large model to generate the first conference minutes data in combination with the preset template. This data includes the voiceprint information, encryption records, and content correspondence of the M identities, and is synchronized to the cloud storage through the APN encryption channel, while the local cache supports temporary storage in the "listen and burn" mode.

[0099] After entering the second period of time, the terminal starts an independent voiceprint recognition process for the newly added N identity list, such as the newly added 3 people: through a high-sensitivity microphone array to collect the voice data of the new personnel, extract the fundamental frequency information and resonance information, and calculate the identity feature value. At this time, the sequence number of the new identity is generated according to the rule of "original maximum sequence number + 1", such as the original sequence numbers 1-10, and the new sequence numbers 11-13, that is, the sorting position is automatically connected as "original M sequence numbers + N new sequence numbers". The terminal generates the second conference minutes data based on the speaking content of the new personnel, and the encryption process reuses the Chinese encoding algorithm and shift rule, such as adjusting the offset value according to the speaking data volume ratio of the new personnel to ensure consistency with the encryption standard of the first minutes.

[0100] This time-sharing sorting mechanism fully adapts to the hardware performance and software architecture of the product: on the one hand, relying on the parallel computing capability of the terminal's high-performance processor, it can simultaneously process the generation of new and old minutes data without affecting the real-time transcription of the meeting, and the performance of supporting 20 conference rooms concurrent processing can guarantee the parallel processing of multi-period data in a single conference room; on the other hand, through the cross-end synchronization function of the cloud conference service platform, the two minutes data can be stored in association in the terminal and external systems (such as OA, party building system), and the user can later merge the two minutes through the terminal interface, and the system will automatically verify the sequence number continuity (1-13 without duplication) and the consistency of the encryption code during the merging process, to ensure the integrity of the meeting records.

[0101] This mechanism effectively deals with the scenario of dynamic adjustment of personnel in meetings, avoiding interference of new personnel with the original encryption process, such as not needing to recalculate the encryption code of the original M identities, and ensuring the accuracy of identity association through sequence number connection. It cooperates with functions such as "automatic transmission of cross-system conference minutes" and "exclusive security encryption" to improve the scalability and flexibility of the system in complex meeting scenarios, and meets the needs of different industries for dynamic meeting management.

[0102] The method further includes the following steps:

[0103] When there are temporary new participants during the meeting, such as risk control experts joining the financial institution meeting halfway or subject leaders temporarily attending the education institution seminar, the intelligent conference terminal can automatically obtain the N new identity lists through real-time data synchronization with external platforms such as enterprise OA systems and party building systems, and trigger the time-sharing sorting mechanism.

[0104] In the first period of time, the terminal relies on the parallel computing power of the high-performance processor to adjust the list numbers of the current existing M identity lists: if the "amplification N+1 times" strategy is adopted; if the original sequence numbers of the M=5 participants are 1-5, and N=4 new people are added, the sequence numbers are adjusted to 5, 10, 15, 20, and 25; or the "shift N positions" strategy is adopted, and the original sequence numbers 1-5 are adjusted to 5-9, providing insertion space for new personnel by reserving sequence number gaps. During this process, the terminal synchronously generates meeting record segments corresponding to the current M identities, completes preliminary encryption in combination with a professional vocabulary library, adopts Chinese encoding algorithm and shift encryption rules, and temporarily stores them in the cloud through APN encryption channel to ensure that the original encryption process is not disturbed by new personnel.

[0105] After entering the second period of time, the terminal starts the voiceprint collection and processing process of the N new identities: the omnidirectional pickup microphone array captures the voice data of the new personnel, extracts the fundamental frequency information and resonance information, calculates the identity characteristic value through the Guangming big model voice small model, and generates the sequence number, such as 1-4 inserted into the reserved gap. After the speech content of the new personnel is converted into text in real time, it is encrypted dynamically and differentially according to its sequence number, the shift offset value is adjusted according to the data volume ratio, and an independent encrypted record segment is formed.

[0106] Finally, the terminal reorders the sequence numbers of the M original identities and the N new identities from small to large, such as merging them into 1-9, calling the template engine in the meeting service, and integrating the encrypted record segments of the two periods of time into complete meeting minutes data. During the integration process, the system checks the data integrity through the SM3 hash algorithm to ensure that the new and old sequence numbers are continuous and non-repetitive, such as 1-9 after recombining 5, 10, 15... and 1-4, and the shift rules of the encrypted code remain consistent. The generated complete minutes can be exported in formats such as Word and PDF, or transmitted to OA, party building and other systems through the "industry and number fusion platform", and also supports enabling the "listen and burn" function to destroy temporary data.

[0107] This mechanism fully utilizes the real-time processing capability of the terminal and the collaborative storage advantage of the cloud, and when dealing with dynamic adjustment of meeting participants, it not only guarantees the stability of the original encrypted data without the need to recalculate the historical code, but also realizes the seamless access of new personnel through sequence number reservation, which is deeply adapted to the product features of "high concurrency processing" and "exclusive security encryption", significantly improving the scalability and encryption flexibility of the system in complex conference scenarios.

[0108] The method further comprises the following steps:

[0109] Before performing the encoding shift encryption operation, the intelligent conference terminal first counts the total number of the list numbers of the current participants, i.e., the sum of M+N, M is the initial number of participants, and N is the number of newly added participants, and judges the parity attribute through logical operation of the terminal processing module. For example, in the enterprise department meeting, if the initial 5 participants (list numbers 1-5) participate without newly added participants, the total number is 5 (odd number), and the first direction offset (such as the encoding value increment direction) is triggered; if the initial 8 participants of the government meeting participate, and 2 participants are newly added, the total number is 10 (even number), and the second direction offset, such as the encoding value decrement direction, is started.

[0110] The mechanism is deeply integrated with the encryption system of the product: the shift rules of the first direction and the second direction are both based on the effective interval design of the Chinese encoding algorithm (such as GB2312), and cooperate with the SM2 and SM3 algorithm system in the exclusive security encryption, to form a multi-layer encryption logic. For example, when the total number is odd, the actual encoding of the word “meeting” is mapped to “nuiyi” after the first direction offset; when the total number is even, the actual encoding of the word “meeting” is mapped to “sheyi” after the second direction offset, and the encryption results of the two directions are not regularly related, which greatly improves the difficulty of unauthorized cracking.

[0111] Meanwhile, the high-performance processor of the terminal will real-time check whether the encrypted encoding after the shift is out of the encoding range, and in combination with the dynamic differentiated encryption strategy, the shift value of the speaker is small, to ensure that the shift in different directions is executed within the safe interval. This dynamic direction adjustment based on the parity of the total number of list numbers fully utilizes the real-time computing capability of the Guangming big model, and is complementary to the “listen and burn” function – the former enhances the anti-cracking property of the encryption process, and the latter eliminates the risk of leakage from the source of the data life cycle, to jointly build a protection system that meets the high-security level conference requirements, and effectively supports the application of the product in the security scene such as finance and government.

[0112] The embodiment of the application further discloses a conference summary automatic generation system driven by an intelligent conference terminal, comprising a processor, and the processor executes the steps of the conference summary automatic generation method driven by the intelligent conference terminal according to any one of the above.

[0113] Although the embodiments of the application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the application.

Claims

1. A method for automatically generating meeting minutes driven by an intelligent conference terminal, characterized in that, Includes the following steps: Obtain the list of participants in the meeting; The system collects audio data from the conference in real time and identifies voiceprint information corresponding to the identity list from the audio data. The voiceprint information includes fundamental frequency information and resonance information. The identity features are calculated based on the fundamental frequency information and the resonance information. The identity features are sorted to obtain a list number that corresponds one-to-one with the identity list. The list number is unique. The identity features are calculated using a weighted average method based on the fundamental frequency information and the resonance information. The weight of the baseband information is adjusted in a positive correlation with the number of identities on the list; the fewer the number of identities on the list, the smaller the weight of the baseband information; the more the number of identities on the list, the larger the weight of the baseband information. The weight of the resonance information is adjusted in a positive correlation with the total amount of sound data; the smaller the total amount of sound data, the smaller the weight of the resonance information; the larger the total amount of sound data, the larger the weight of the resonance information. The audio data is recognized as meeting record data with text content, and the identity list is associated with the meeting record data; The text content in the meeting record data is converted into actual encoding based on a Chinese encoding algorithm with a continuous encoding range; Based on the association, the identity list is traversed, and the actual encoding of the meeting record data is shifted according to the list number corresponding to the identity list to obtain the encrypted encoding; The corresponding encrypted record is matched from the Chinese encoding algorithm according to the encryption encoding; Based on the association, the content correspondence between the list serial number and the encrypted record is obtained; The identity list, the list number, the encrypted record, and the content correspondence are stored as meeting minutes data; The parity attribute is obtained based on the total number of the sequence numbers in the list; If the parity attribute is odd, the shift is along the first direction; otherwise, the shift is along the second direction. If N new identities are added to the list during the generation of the meeting record data, then the data is sorted by time. During a first time period, the sequence number of the current M identity lists is generated, wherein the sorting position of the M identity lists is magnified by N+1 times; During the second time period, generate the list numbers of the N newly added identity lists; Insert the current M identity lists into the newly added N identity lists, reorder them according to the corresponding list numbers, and then generate the meeting minutes data.

2. The method for automatically generating meeting minutes driven by an intelligent conference terminal according to claim 1, characterized in that, The method also includes the following steps: Obtain the meeting minutes data; Extract the identity list, the list number, the encrypted record, and the content correspondence from the meeting minutes data; The encrypted code is matched based on the encrypted record and the reverse algorithm of the Chinese encoding algorithm; The list number is matched with the encryption code according to the correspondence between the encrypted record and the content; The actual code is obtained by reverse-shifting the encrypted code according to the corresponding list number; The meeting record data is matched based on the actual encoding and the reverse algorithm of the Chinese encoding algorithm; Convert the content correspondence into the association relationship; Output the identity list, the association relationships, and the meeting record data.

3. The method for automatically generating meeting minutes driven by a smart conference terminal according to claim 1 or 2, characterized in that, The method also includes the following steps: If any two corresponding identity lists have the same identity feature value, then compare the values ​​of the baseband information corresponding to the two identity features, and increase the value of the identity feature with the larger baseband information value; If any two corresponding identity lists have the same identity feature value and the corresponding fundamental frequency information value is also the same, then compare the values ​​of the resonance information corresponding to the two features and increase the value of the identity feature with the larger resonance information value.

4. The method for automatically generating meeting minutes driven by an intelligent conference terminal according to claim 1, characterized in that, The method also includes the following steps: The shift offset value of the identity list is adjusted inversely based on the proportion of the text content corresponding to the identity list in the meeting record data; the smaller the proportion of the data, the larger the shift offset value of the identity list; the larger the proportion of the data, the smaller the shift offset value of the identity list.

5. The method for automatically generating meeting minutes driven by an intelligent conference terminal according to claim 1, characterized in that, The method also includes the following steps: If the identity list is added during the generation of the meeting record data, then the data will be sorted by time. Generate a meeting minutes data corresponding to the current M identity lists within a first time period; In the second period, another meeting minutes data corresponding to the newly added N identity lists is generated, wherein the sorting position of the newly added list number is +M.

6. A meeting minutes automatic generation system driven by an intelligent conference terminal, characterized in that, The system includes a processor that performs the steps of the automatic generation method for meeting minutes driven by a smart conference terminal as described in any one of claims 1-5.

Images