A multi-party intercom intelligent call method and system

Abstract

The present invention relates to the field of intelligent call technology, and discloses a multi-party intercom intelligent call method and system. The multi-party intercom intelligent call method and system can obtain higher quality input data through independent sound source labeling and multi-source data collection, thereby improving the accuracy of subsequent processing. Secondly, the structured process of the method in data processing, calculation and analysis makes the noise reduction judgment more accurate and intelligent, avoiding the problems of misjudgment and misprocessing in traditional methods. Finally, the flexible execution strategy of the method ensures the maximization of the noise reduction effect and improves the overall sound quality and user experience of the call. Compared with traditional call noise reduction technology, the method has not only been improved in accuracy and efficiency, but also enables the method to better cope with complex call environments, significantly improving the effect and experience of multi-party calls.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent conversation, and in particular to a multi-party intercom intelligent conversation method and system. Background Art

[0002] In the wide application of modern communication technology, multi-party intercom systems are regarded as key real-time communication tools and are widely used in emergency response, teamwork, remote collaboration and other scenarios. With the diversification of communication needs, how to maintain high-quality call effects in complex environments has become the focus of research and development.

[0003] However, traditional multi-party intercom systems often have difficulty coping with multiple interferences and noises in complex and changing communication environments. Existing noise reduction technologies usually rely on a single algorithm model, which makes it difficult to fully consider the diversity and dynamic changes of various environmental factors. In addition, these systems usually ignore the personalized voice characteristics of the participants and cannot be flexibly adjusted according to the characteristics of different sound sources. This leads to the fact that in practical applications, traditional methods are prone to misjudgment, such as failure to effectively filter noise or incorrect processing of voice signals, which affects the clarity of the call and user experience; therefore, it does not meet existing needs. In this regard, we propose a multi-party intercom intelligent call method. Summary of the invention

[0004] The present invention provides a multi-party intercom intelligent call method, which is used in the traditional multi-party intercom system mentioned in the above background technology. It is often difficult to cope with various interferences and noises in a complex and changeable communication environment. Existing noise reduction technologies usually rely on a single algorithm model, which is difficult to fully consider the diversity and dynamic changes of various environmental factors. In addition, these systems usually ignore the personalized voice characteristics of the participants and cannot be flexibly adjusted according to the characteristics of different sound sources. This leads to the traditional methods being prone to misjudgment in practical applications, such as being unable to effectively filter noise or incorrectly processing voice signals, thereby affecting the clarity of the call and the user experience.

[0005] The present invention provides the following technical solution: a multi-party intercom intelligent call method, characterized in that: the specific steps are as follows:

[0006] S1, data collection, marking each sound source in the call, and independently collecting multi-source data under each sound source according to the marking;

[0007] S2, data processing, processing the collected multi-source data, and arranging the processed multi-source data into a first data set, a second data set, a third data set, and a fourth data set;

[0008] S3, data calculation, integrating and calculating the first data set, the second data set, the third data set and the fourth data set, so as to generate a recognition coefficient BSX;

[0009] S4, data analysis, comparing the calculated recognition coefficient with a preset first threshold value Y, thereby generating a first comparison result, and judging whether the current sound source needs to be denoised according to the first comparison result;

[0010] S5. Data analogy: if the first comparison result indicates that noise reduction is required, the identification coefficient BSX is calculated with the preset first threshold Y to generate a magnitude coefficient LJZ, and the magnitude coefficient LJZ is compared with the preset second threshold R to generate a second comparison result;

[0011] S6. Execute, and perform noise reduction according to the second comparison result.

[0012] Preferably, in step S1, each sound source in the call is marked as i1, i2, i3, ..., in respectively;

[0013] The multi-source data for each sound includes an audio signal dataset, an environmental noise dataset, a participant feature dataset, and a correction dataset;

[0014] The audio signal data set includes microphone input signal strength, spectrum waveform, time domain waveform change rate and speech energy ratio;

[0015] The environmental noise data set includes the background noise amount and noise spectrum distribution rate;

[0016] The participant characteristics dataset included vocalization frequency range, utterance duration, speaking volume, and speech rate;

[0017] The modified data set includes real-time noise change rate, inter-device delay, and reverberation quality.

[0018] Preferably, in step S2, the multi-source data collected in step S1 are preprocessed, the preprocessed multi-source data are dimensionless, and the dimensionless multi-source data are reorganized to generate a first data set, a second data set, a third data set, and a fourth data set;

[0019] The first data set includes microphone input signal strength, spectrum distribution rate, time domain waveform change rate and speech energy ratio;

[0020] The microphone input signal strength is recorded as MKQ according to the sound source label. i1 MKQ i2 MKQ i3 , ..., MKQ in ;

[0021] The spectrum waveform is marked as PPF according to the sound source i1 、PPF i2、PPF i3 、...、PPF in ;

[0022] The rate of change of the time domain waveform is denoted by the sound source label SBX i1 , SBX i2 , SBX i3 ,...,SBX in ;

[0023] The speech energy ratio is denoted as YLB according to the sound source label. i1 , YLB i2 , YLB i3 、...、YLB in ;

[0024] The second data set includes the amount of background noise and the noise spectrum distribution ratio;

[0025] The background noise is denoted as BJZ according to the sound source label. i1 , BJZ i2 , BJZ i3 、...、BJZ in ;

[0026] The noise spectrum distribution rate is recorded as ZPP according to the sound source label i1 , ZPP i2 , ZPP i3 ,..., ZPP in ;

[0027] The third dataset includes the vocalization frequency range, utterance duration, speaking volume, and speech rate;

[0028] The sound frequency range is marked as FSL according to the sound source. i1 、FSL i2 、FSL i3 、...、FSL in ;

[0029] The duration of speech is recorded as YSC according to the sound source mark i1 , YSC i2 , YSC i3 、...、YSC in ;

[0030] The speaking volume is recorded as HYL according to the sound source label. i1 HYL i2 HYL i3 ,...,HYL in ;

[0031] The speaking rate is recorded as HYS according to the sound source marker. i1HYS i2 HYS i3 ,...,HYS in ;

[0032] The fourth data set includes real-time noise change rate, inter-device delay, and reverberation quality;

[0033] The real-time noise change rate is denoted by the sound source label as SZB i1 , SZB i2 , SZB i3 ,..., SZB in ;

[0034] The delay between devices is recorded as YCS according to the sound source label i1 、YCS i2 、YCS i3 ,...,YCS in ;

[0035] The reverberation quality is marked as HXB according to the sound source. i1 ,HXB i2 ,HXB i3 、...、HXB in .

[0036] Preferably, in step S3, the identification coefficient BSX is calculated and obtained by the following formula:

[0037] ;

[0038] Wherein: S1, S2 and S3 are the first reference coefficient, the second reference coefficient and the third reference coefficient respectively, K is the intervention reference coefficient, the values ​​of S1, S2, S3 and K are calculated by integrating the first data set, the second data set, the third data set and the fourth data set respectively, a1, a2 and a3 are weight values, and the values ​​of a1, a2 and a3 are adjusted and set by the user.

[0039] Preferably, the first reference coefficient S1, the second reference coefficient S2 and the third reference coefficient S3 are respectively calculated and obtained by the following formulas:

[0040] ;

[0041]

[0042] ;

[0043] In the formula: MKQ in 、PPF in , SBX in , YLB in , BJZ in, ZPP in 、FSL in , YSC in HYL in HYS in They are the microphone input signal strength, spectrum waveform, time domain waveform change rate, speech energy ratio, background noise amount, noise spectrum distribution rate, vocalization frequency range, speech duration, speaking volume and speaking speed under the sound source marked as in.

[0044] Preferably, the intervention reference value K is calculated and obtained by the following formula:

[0045] ;

[0046] Where: SZB in 、YCS in and HXB in They are respectively the real-time noise change rate, inter-device delay and reverberation sound quality under the sound source marked as in, b1 and b2 are weight values, and the values ​​of b1 and b2 are adjusted and set by the user.

[0047] Preferably, in step S4, the first comparison result is as follows:

[0048] when When , it means the current sound source is normal and no noise reduction is needed;

[0049] when , it means the current sound source is abnormal and noise reduction is required.

[0050] Preferably, in step S5, the second comparison result is as follows:

[0051] when When , it means the current sound source is level 1 abnormal;

[0052] when , it means the current sound source is a level 2 abnormality.

[0053] Preferably, the magnitude coefficient in step S5 is calculated and obtained by the following formula:

[0054] ;

[0055] Where: BSX is the identification coefficient, Y is the first threshold.

[0056] The present application also includes a multi-party intercom intelligent call system, the system includes a data acquisition unit, a data processing unit, a data calculation unit, a data analysis unit and an execution unit;

[0057] A data collection unit, used to mark each sound source in the call and independently collect multi-source data of each sound source according to the mark;

[0058] A data processing unit, used to process the collected multi-source data and organize the processed multi-source data into a first data set, a second data set, a third data set and a fourth data set;

[0059] A data calculation unit, used for integrating and calculating the first data set, the second data set, the third data set and the fourth data set, so as to generate a recognition coefficient BSX;

[0060] A data analysis unit, configured to compare the obtained identification coefficient with a preset first threshold value Y to generate a first comparison result, and to calculate the identification coefficient BSX with the preset first threshold value Y to generate a magnitude coefficient LJZ, and to compare the magnitude coefficient LJZ with a preset second threshold value R;

[0061] The execution unit is used to perform noise reduction according to the second comparison result.

[0062] The present invention has the following beneficial effects:

[0063] 1. This multi-party intercom intelligent call method can obtain higher quality input data through independent sound source labeling and multi-source data collection, thereby improving the accuracy of subsequent processing. Secondly, the structured process of this method in data processing, calculation and analysis makes the noise reduction judgment more accurate and intelligent, avoiding the problems of misjudgment and misprocessing existing in traditional methods. Finally, the flexible execution strategy of this method ensures the maximization of the noise reduction effect and improves the overall sound quality and user experience of the call. Compared with traditional call noise reduction technology, this method has not only improved the accuracy and efficiency, but also enables this method to better cope with complex call environments, significantly improving the effect and experience of multi-party calls.

[0064] 2. This multi-party intercom intelligent call method integrates reference coefficients S1, S2, and S3 from different data sets, and the calculation of the recognition coefficient BSX can fully reflect the various factors in the call environment. The combination of different reference coefficients takes into account the characteristics of the audio signal, the characteristics of the environmental noise, and the personalized voice characteristics of the participants, so that the recognition coefficient BSX can accurately reflect the status and needs of each sound source. The introduction of the intervention reference coefficient K ensures the consistency and accuracy of the calculation results by standardizing and calibrating the overall calculation process. Whether it is processing audio data from different sound sources or facing a complex noise environment, K can effectively balance the impact of various factors on the recognition coefficient BSX, thereby improving the accuracy of the recognition coefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a step diagram of the method of the present invention.

[0066] Figure 2 It is a flow chart of the system of the present invention. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] Example 1

[0069] See also Figure 1 , a multi-party intercom intelligent call method, characterized in that: the specific steps are as follows:

[0070] S1, data collection, marking each sound source in the call, and independently collecting multi-source data under each sound source according to the marking;

[0071] S2, data processing, processing the collected multi-source data, and arranging the processed multi-source data into a first data set, a second data set, a third data set, and a fourth data set;

[0072] S3, data calculation, integrating and calculating the first data set, the second data set, the third data set and the fourth data set, so as to generate a recognition coefficient BSX;

[0073] S4, data analysis, comparing the calculated recognition coefficient with a preset first threshold value Y, thereby generating a first comparison result, and judging whether the current sound source needs to be denoised according to the first comparison result;

[0074] S5. Data analogy: if the first comparison result indicates that noise reduction is required, the identification coefficient BSX is calculated with the preset first threshold Y to generate a magnitude coefficient LJZ, and the magnitude coefficient LJZ is compared with the preset second threshold R to generate a second comparison result;

[0075] S6. Execute, and perform noise reduction according to the second comparison result.

[0076] In this embodiment: In step S1, by marking each sound source in the call, the voice of each participant can be accurately identified and distinguished. This independent sound source marking method ensures that different sound sources will not be confused when collecting data, thereby laying a solid foundation for subsequent noise reduction processing. By independently collecting multi-source data (such as audio signals, environmental noise, participant characteristics, etc.) under each sound source, this method can obtain rich and diverse data, providing comprehensive input for accurate sound processing and analysis. This process effectively avoids interference and errors in data collection and ensures the high quality and pertinence of the collected data.

[0077] In step S2, the collected multi-source data is deeply analyzed and classified. By arranging the processed data into a first data set, a second data set, a third data set, and a fourth data set, the method can better manage and analyze different types of data. This classification process not only improves the efficiency of data management, but also provides structured input for subsequent data calculations, making the calculation process more organized and accurate. At the same time, this step effectively completes the preprocessing and cleaning of the data, laying a good foundation for the subsequent noise reduction process.

[0078] In step S3, the first data set, the second data set, the third data set and the fourth data set are integrated and calculated to generate the identification coefficient BSX. Through the integrated calculation, the method can extract key information from multiple data sources and convert it into an identification coefficient that can reflect the characteristics of the sound source. The generation of this coefficient is a key link in the subsequent judgment of whether the sound source needs noise reduction, ensuring that the method can take appropriate noise reduction measures according to different sound source conditions. Through the precise calculation process, the method completes the fusion and analysis of multi-source data, and improves the accuracy and reliability of subsequent judgments.

[0079] In step S4, the first comparison result is generated by comparing the identification coefficient BSX with the preset first threshold value Y. The core of this step is to analyze the sound source characteristics to determine whether the current sound source needs noise reduction. By comparing with the preset threshold, the method can effectively distinguish between the sound source with large noise and need to be processed and the sound source with small noise and can be ignored. This step ensures the pertinence of the noise reduction processing of the method, avoids unnecessary noise reduction operations, and thus optimizes the overall performance and resource utilization efficiency of the method.

[0080] In step S5, if noise reduction is required, the magnitude coefficient LJZ is further calculated and compared with the preset second threshold R to generate a second comparison result. By calculating the magnitude coefficient, the method can evaluate the severity of the noise and determine an appropriate noise reduction strategy accordingly. By comparing the magnitude coefficient with the second threshold, the method can more accurately judge the necessity and urgency of noise reduction and ensure that the effect of noise reduction processing is maximized. This step makes the method more flexible and intelligent when performing noise reduction, and can make the best processing decision according to the specific situation.

[0081] In step S6, according to the second comparison result, the method performs a noise reduction operation. The accuracy and effect of the noise reduction process directly depend on the analysis and calculation results of the previous steps. Through accurate judgment and flexible processing strategies, the method can effectively reduce noise interference during the call, thereby improving the sound quality of the call and the experience of the participants. This step marks the final completion of the entire intelligent call noise reduction process, achieving the expected noise reduction goal, making the call clearer and smoother.

[0082] Through independent sound source labeling and multi-source data collection, this method can obtain higher quality input data, thereby improving the accuracy of subsequent processing. Secondly, the structured process of this method in data processing, calculation and analysis makes the noise reduction judgment more accurate and intelligent, avoiding the misjudgment and misprocessing problems existing in traditional methods. Finally, the flexible execution strategy of this method ensures the maximization of the noise reduction effect and improves the overall sound quality and user experience of the call. Compared with traditional call noise reduction technology, this method has not only improved the accuracy and efficiency, but also enabled this method to better cope with complex call environments, significantly improving the effect and experience of multi-party calls.

[0083] Example 2

[0084] See also Figure 1 ,In step S1, each sound source in the call is marked as i1, i2, i3, ..., in;

[0085] The multi-source data for each sound includes an audio signal dataset, an environmental noise dataset, a participant feature dataset, and a correction dataset;

[0086] The audio signal data set includes microphone input signal strength, spectrum waveform, time domain waveform change rate and speech energy ratio;

[0087] The environmental noise data set includes the background noise amount and noise spectrum distribution rate;

[0088] The participant characteristics dataset included vocalization frequency range, utterance duration, speaking volume, and speech rate;

[0089] The modified data set includes real-time noise change rate, inter-device delay, and reverberation quality.

[0090] In this embodiment: By marking each sound source and collecting a comprehensive and detailed multi-source data set, this method has achieved significant improvements in many aspects. First, independent labeling and data collection ensure the accurate identification and processing of each sound source, avoiding confusion problems that may occur in traditional methods. Secondly, this method provides a customized noise reduction solution for each participant through a sophisticated audio signal data set, an environmental noise data set, a participant feature data set, and a correction data set. This not only improves voice clarity and call quality, but also enables the method to more flexibly deal with special situations such as complex noise environments and equipment delays. Overall, this multi-party intercom intelligent call method has achieved comprehensive improvements in accuracy, flexibility, and user experience, and is significantly superior to traditional technical means.

[0091] Example 3

[0092] See also Figure 1 In step S2, the multi-source data collected in step S1 are preprocessed, the preprocessed multi-source data are dimensionless, and the dimensionless multi-source data are reorganized to generate a first data set, a second data set, a third data set, and a fourth data set;

[0093] The first data set includes microphone input signal strength, spectrum distribution rate, time domain waveform change rate and speech energy ratio;

[0094] The microphone input signal strength is recorded as MKQ according to the sound source label. i1 MKQ i2 MKQ i3 , ..., MKQ in ;

[0095] The spectrum waveform is marked as PPF according to the sound source i1 、PPF i2 、PPF i3 、...、PPF in ;

[0096] The rate of change of the time domain waveform is denoted by the sound source label SBX i1 , SBX i2 , SBX i3 ,...,SBX in ;

[0097] The speech energy ratio is denoted as YLB according to the sound source label. i1 , YLB i2 , YLB i3 、...、YLB in ;

[0098] The second data set includes the amount of background noise and the noise spectrum distribution ratio;

[0099] The background noise is denoted as BJZ according to the sound source label. i1 , BJZ i2 , BJZ i3 、...、BJZ in ;

[0100] The noise spectrum distribution rate is recorded as ZPP according to the sound source label i1 , ZPP i2 , ZPP i3 ,..., ZPP in ;

[0101] The third dataset includes the vocalization frequency range, utterance duration, speaking volume, and speech rate;

[0102] The sound frequency range is marked as FSL according to the sound source. i1 、FSL i2 、FSL i3 、...、FSL in ;

[0103] The duration of speech is recorded as YSC according to the sound source mark i1 , YSC i2 , YSC i3 、...、YSC in ;

[0104] The speaking volume is recorded as HYL according to the sound source label. i1 HYL i2 HYL i3 ,...,HYL in ;

[0105] The speaking rate is recorded as HYS according to the sound source marker. i1 HYS i2 HYS i3 ,...,HYS in ;

[0106] The fourth data set includes real-time noise change rate, inter-device delay, and reverberation quality;

[0107] The real-time noise change rate is denoted by the sound source label as SZB i1 , SZB i2 , SZB i3 ,..., SZB in ;

[0108] The delay between devices is recorded as YCS according to the sound source label i1 、YCS i2 、YCS i3,...,YCS in ;

[0109] The reverberation quality is marked as HXB according to the sound source. i1 ,HXB i2 ,HXB i3 、...、HXB in .

[0110] In this embodiment: by sorting out the first data set including microphone input signal strength, spectrum distribution rate, time domain waveform change rate and speech energy ratio, accurate speech feature information is provided, which greatly improves the effects of speech processing and noise reduction.

[0111] By collating the second data set including the background noise amount and the noise spectrum distribution rate, and isolating these noise-related data, the system can perform special analysis and processing on the noise, thereby more effectively reducing the impact of background noise on call quality.

[0112] The third data set includes the range of vocalization frequencies, speech duration, speaking volume, and speech speed. This data set covers the personalized voice characteristics of the participants, allowing the system to customize processing based on the different voice characteristics of each participant, improving the naturalness and personalized experience of the call.

[0113] The fourth data set includes real-time noise change rate, inter-device delay, and reverberation quality. The setting of this data set enables the system to dynamically adjust and optimize processing parameters to ensure a high-quality call experience in a complex and changing environment.

[0114] By preprocessing and dimensionlessizing multi-source data and organizing it into four targeted data sets, the system has significantly improved the accuracy, flexibility and applicability of data processing. Compared with traditional processing methods, this multi-step and sophisticated processing process ensures that all types of data can be analyzed and calculated in the most suitable form, thereby achieving higher voice clarity, personalized processing and environmental adaptability. This system improvement not only improves the quality of voice calls, but also greatly improves the user experience, allowing the system to provide stable and high-quality services in complex and changing environments.

[0115] Example 4

[0116] See also Figure 1 In step S3, the identification coefficient BSX is calculated by the following formula:

[0117] ;

[0118] Wherein: S1, S2 and S3 are the first reference coefficient, the second reference coefficient and the third reference coefficient respectively, K is the intervention reference coefficient, the values ​​of S1, S2, S3 and K are calculated by integrating the first data set, the second data set, the third data set and the fourth data set respectively, a1, a2 and a3 are weight values, and the values ​​of a1, a2 and a3 are adjusted and set by the user.

[0119] In this embodiment: by integrating reference coefficients S1, S2, and S3 from different data sets, the calculation of the recognition coefficient BSX can fully reflect various factors in the call environment. The combination of different reference coefficients takes into account the audio signal characteristics, the environmental noise characteristics, and the personalized voice characteristics of the participants, so that the recognition coefficient BSX can accurately reflect the status and requirements of each sound source.

[0120] Users can control the influence of different reference coefficients in the calculation of the identification coefficient BSX by adjusting the weights a1, a2 and a3. This flexibility allows the system to optimize according to the actual situation of different call scenarios, such as increasing the weight of S2 in a high-noise environment to pay more attention to the impact of noise on the call, or increasing the weight of S3 in a multi-person intercom to better adapt to the voice differences of the participants.

[0121] The introduction of the intervention reference coefficient K ensures the consistency and accuracy of the calculation results by standardizing and calibrating the overall calculation process. Whether processing audio data from different sound sources or facing complex noise environments, K can effectively balance the impact of various factors on the recognition coefficient BSX, thereby improving the accuracy of the recognition coefficient.

[0122] By calculating the identification coefficient BSX, the system can more accurately determine which sound sources need to be processed for noise reduction. This not only improves the effectiveness of noise reduction processing, but also reduces unnecessary processing, avoids waste of resources, and optimizes call quality in multi-party intercom.

[0123] The calculation method of the identification coefficient BSX integrates multi-source data and allows users to adjust the weights to adapt to different call needs. Compared with traditional noise reduction technology, this calculation method based on the integration of multiple data sets and customized weights greatly improves the accuracy, flexibility and adaptability of the system, allowing the system to maintain excellent call quality in complex call environments.

[0124] Example 5

[0125] See also Figure 1 , the first reference coefficient S1, the second reference coefficient S2 and the third reference coefficient S3 are respectively calculated and obtained by the following formulas;

[0126] ;

[0127]

[0128] ;

[0129] In the formula: MKQ in 、PPF in , SBX in , YLB in , BJZ in , ZPP in 、FSL in , YSC in HYL in HYS in They are respectively the microphone input signal strength, spectrum waveform, time domain waveform change rate, speech energy ratio, background noise amount, noise spectrum distribution rate, vocalization frequency range, speech duration, speaking volume and speaking speed under the sound source marked as in, and e is the base function.

[0130] The intervention reference value K is calculated by the following formula:

[0131] ;

[0132] Where: SZB in 、YCS in and HXB in They are the real-time noise change rate, inter-device delay and reverberation quality under the sound source marked as in, b1 and b2 are weight values, and the values ​​of b1 and b2 are adjusted and set by the user, and ln is a logarithmic function.

[0133] In this embodiment, the microphone input signal strength is combined with the spectrum waveform, the time domain waveform change rate, and the speech energy ratio through the calculation formula of the first reference coefficient S1 to effectively capture and characterize the signal characteristics of different sound sources. In this way, multiple sound sources can be distinguished more accurately, thereby providing higher accuracy for subsequent noise reduction decisions.

[0134] The second reference coefficient S2 formula uses the exponential form of the background noise amount and the noise spectrum distribution rate to capture the nonlinear changes in noise characteristics. This enables the system to more sensitively identify and adjust the processing strategy when facing a complex and dynamically changing noise environment, thereby improving the clarity and stability of the call.

[0135] The third reference coefficient S3 formula deeply analyzes the voice characteristics of the participants through the relationship between the vocalization frequency range and the duration of the speech, speaking volume and speaking speed. This personalized feature analysis helps to optimize the voice processing in the call, so that the voice characteristics of each participant can be fully recognized and retained, avoiding distortion or loss of information during the noise reduction process.

[0136] By calculating the first reference coefficient S1, the second reference coefficient S2 and the third reference coefficient S3, the system can comprehensively and accurately capture the characteristics of each sound source in the call, and combine nonlinear noise analysis and personalized voice feature analysis to greatly improve the accuracy of voice recognition, the system's adaptability to noisy environments, and the processing effect of participants' voices. Compared with traditional single or static noise reduction processing methods, this algorithm based on multi-dimensional and dynamic characteristic analysis significantly improves the voice quality and user experience in multi-party intercoms.

[0137] The calculation formula of the intervention reference value K combines the influence of two important parameters: inter-device delay and reverberation sound quality. This allows the system to consider not only the environmental noise but also the performance of the device itself when evaluating call quality. In this way, the system can more accurately identify and handle sound quality degradation caused by equipment problems, thereby improving the overall call quality. Users can adjust the weight values ​​of b1 and b2 according to actual needs to meet the noise reduction needs in different scenarios. For example, in a high-noise environment, the weight of b1 can be increased to emphasize the impact of the noise change rate; in the case of large inter-device delays, the weight of b2 can be increased to enhance the ability to handle equipment problems. This flexibility allows the system to provide optimal call quality in different application scenarios.

[0138] The calculation formula of the intervention reference value K achieves more accurate and dynamic sound quality processing by comprehensively considering noise changes, equipment performance and user control. Compared with the traditional static noise reduction method, this method has significantly improved the real-time adaptability, accuracy of sound quality processing and personalized adjustment, thus greatly improving the call quality and user experience of the multi-party intercom system.

[0139] Example 6

[0140] See also Figure 1 , in step S4, the first comparison result is as follows;

[0141] when When , it means the current sound source is normal and no noise reduction is needed;

[0142] when , it means the current sound source is abnormal and noise reduction is required.

[0143] In step S5, the second comparison result is as follows;

[0144] when When , it means the current sound source is level 1 abnormal;

[0145] when , it means the current sound source is a level 2 abnormality.

[0146] In step S5, the magnitude coefficient is calculated by the following formula:

[0147] ;

[0148] Where: BSX is the identification coefficient, Y is the first threshold.

[0149] In this embodiment: by comparing the recognition coefficient BSX with the first threshold value Y, the system can quickly determine whether the current sound source is normal. , it indicates that the sound source is in good condition and no further processing is required. This fast judgment mechanism helps reduce unnecessary noise reduction processing, thereby saving system resources and improving response efficiency. The setting of the first threshold Y ensures that the noise reduction process is triggered only when the sound source is significantly abnormal. This mechanism effectively reduces the possibility of misjudgment and avoids excessive processing of normal sound sources, thereby maintaining the naturalness of the call and the original state of the sound quality.

[0150] By calculating the magnitude coefficient LJZ and comparing it with the second threshold R, the system can further refine the abnormality of the sound source. When , the system identifies it as a first-level abnormality and takes a milder noise reduction approach; when , it is identified as a secondary abnormality and a stronger noise reduction process is performed. This hierarchical processing strategy allows the system to implement targeted noise reduction measures according to the severity of the abnormality, thereby effectively eliminating noise while maintaining sound quality. The calculation formula of the magnitude coefficient LJZ comprehensively considers the difference between the recognition coefficient BSX and the first threshold Y, as well as the fluctuation range of the recognition coefficient itself. This comprehensive calculation method ensures that when the system processes sound source abnormalities, it can adaptively adjust the processing intensity according to the actual situation, providing a more flexible and sophisticated noise reduction effect.

[0151] Through the flexible use of the recognition coefficient and magnitude coefficient in steps S4 and S5, the system achieves efficient and accurate sound source status evaluation and exception handling. Compared with the traditional single noise reduction processing method, this processing method based on multi-level evaluation can more effectively reduce the misjudgment rate, provide targeted noise reduction processing, and significantly improve the call quality and user experience of the multi-party intercom system. The system not only has strong real-time adaptability, but also can flexibly respond to different degrees of noise interference, bringing users a clearer and more natural call experience.

[0152] See also Figure 2 , the present application also includes a multi-party intercom intelligent call system, the system includes a data acquisition unit, a data processing unit, a data calculation unit, a data analysis unit and an execution unit;

[0153] A data collection unit, used to mark each sound source in the call and independently collect multi-source data of each sound source according to the mark;

[0154] A data processing unit, used to process the collected multi-source data and organize the processed multi-source data into a first data set, a second data set, a third data set and a fourth data set;

[0155] A data calculation unit, used for integrating and calculating the first data set, the second data set, the third data set and the fourth data set, so as to generate a recognition coefficient BSX;

[0156] A data analysis unit, configured to compare the obtained identification coefficient with a preset first threshold value Y to generate a first comparison result, and to calculate the identification coefficient BSX with the preset first threshold value Y to generate a magnitude coefficient LJZ, and to compare the magnitude coefficient LJZ with a preset second threshold value R;

[0157] The execution unit is used to perform noise reduction according to the second comparison result.

[0158] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0159] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A multi-party intercom intelligent conversation method, characterized by: The specific steps are as follows: S1, data collection, marking each sound source in the call, and independently collecting multi-source data from each sound source according to the marking; In step S1, each sound source in the call is marked as i1, i2, i3, ..., in; The multi-source data for each sound includes an audio signal dataset, an environmental noise dataset, a participant feature dataset, and a correction dataset; The audio signal data set includes microphone input signal strength, spectrum waveform, time domain waveform change rate and speech energy ratio; The environmental noise data set includes the background noise amount and noise spectrum distribution rate; The participant characteristics dataset included vocalization frequency range, utterance duration, speaking volume, and speech rate; The correction data set includes real-time noise change rate, inter-device delay, and reverberation quality; S2, data processing, processing the collected multi-source data, and arranging the processed multi-source data into a first data set, a second data set, a third data set, and a fourth data set; In step S2, the multi-source data collected in step S1 are preprocessed, the preprocessed multi-source data are dimensionless, and the dimensionless multi-source data are reorganized to generate a first data set, a second data set, a third data set, and a fourth data set; The first data set includes microphone input signal strength, spectrum distribution rate, time domain waveform change rate and speech energy ratio; The microphone input signal strength is recorded as MKQ according to the sound source label. i1 MKQ i2 MKQ i3 , ..., MKQ in ; The spectrum waveform is marked as PPF according to the sound source i1 、PPF i2 、PPF i3 、...、PPF in ; The rate of change of the time domain waveform is denoted by the sound source label SBX i1 , SBX i2 , SBX i3 ,...,SBX in ; The speech energy ratio is denoted as YLB according to the sound source label. i1 , YLB i2 , YLB i3 、...、YLB in ; The second data set includes the amount of background noise and the noise spectrum distribution ratio; The background noise is denoted as BJZ according to the sound source label. i1 , BJZ i2 , BJZ i3 、...、BJZ in ; The noise spectrum distribution rate is recorded as ZPP according to the sound source label i1 , ZPP i2 , ZPP i3 ,..., ZPP in ; The third dataset includes the vocalization frequency range, utterance duration, speaking volume, and speech rate; The sound frequency range is marked as FSL according to the sound source. i1 、FSL i2 、FSL i3 、...、FSL in ; The duration of speech is recorded as YSC according to the sound source mark i1 , YSC i2 , YSC i3 、...、YSC in ; The speaking volume is recorded as HYL according to the sound source label. i1 HYL i2 HYL i3 ,...,HYL in ; The speaking rate is recorded as HYS according to the sound source marker. i1 HYS i2 HYS i3 ,...,HYS in ; The fourth data set includes real-time noise change rate, inter-device delay, and reverberation quality; The real-time noise change rate is denoted by the sound source label as SZB i1 , SZB i2 , SZB i3 ,..., SZB in ; The delay between devices is recorded as YCS according to the sound source label i1 、YCS i2 、YCS i3 ,...,YCS in ; The reverberation quality is marked as HXB according to the sound source. i1 ,HXB i2 ,HXB i3 、...、HXB in ; S3, data calculation, integrating and calculating the first data set, the second data set, the third data set and the fourth data set, so as to generate a recognition coefficient BSX; In step S3, the identification coefficient BSX is calculated and obtained by the following formula: Wherein: S1, S2 and S3 are the first reference coefficient, the second reference coefficient and the third reference coefficient respectively, K is the intervention reference coefficient, the values ​​of S1, S2, S3 and K are calculated and obtained by integrating the first data set, the second data set, the third data set and the fourth data set respectively, a1, a2 and a3 are weight values, and the values ​​of a1, a2 and a3 are adjusted and set by the user; The first reference coefficient S1, the second reference coefficient S2 and the third reference coefficient S3 are respectively calculated and obtained by the following formulas; In the formula: MKQ in 、PPF in , SBX in , YLB in , BJZ in , ZPP in 、FSL in , YSC in HYL in HYS in They are the microphone input signal strength, spectrum waveform, time domain waveform change rate, speech energy ratio, background noise amount, noise spectrum distribution rate, utterance frequency range, speech duration, speaking volume and speaking speed under the sound source marked as in; The intervention reference value K is calculated by the following formula: Where: SZB in 、YCS in and HXB in They are the real-time noise change rate, inter-device delay, and reverberation quality under the sound source marked as in, b1 and b2 are weight values, and the values ​​of b1 and b2 are adjusted and set by the user; S4, data analysis, comparing the calculated recognition coefficient with a preset first threshold value Y, thereby generating a first comparison result, and judging whether the current sound source needs to be denoised according to the first comparison result; The first comparison results are as follows; when When , it means the current sound source is normal and no noise reduction is needed; when When , it means the current sound source is abnormal and noise reduction is needed; S5. Data analogy: if the first comparison result indicates that noise reduction is required, the identification coefficient BSX is calculated with the preset first threshold Y to generate a magnitude coefficient LJZ, and the magnitude coefficient LJZ is compared with the preset second threshold R to generate a second comparison result; The second comparison results are as follows; when When , it means the current sound source is level 1 abnormal; when When , it means the current sound source is a level 2 abnormality; The magnitude coefficient LJZ is calculated by the following formula: Where: BSX is the identification coefficient, Y is the first threshold; S6. Execute, and perform noise reduction according to the second comparison result.

2. A multi-party intercom intelligent call system, used to implement the multi-party intercom intelligent call method described in claim 1, characterized in that: The system includes a data acquisition unit, a data processing unit, a data calculation unit, a data analysis unit and an execution unit; A data collection unit, used to mark each sound source in the call and independently collect multi-source data of each sound source according to the mark; A data processing unit, used to process the collected multi-source data and organize the processed multi-source data into a first data set, a second data set, a third data set and a fourth data set; A data calculation unit, used for integrating and calculating the first data set, the second data set, the third data set and the fourth data set, so as to generate a recognition coefficient BSX; A data analysis unit, configured to compare the obtained identification coefficient with a preset first threshold value Y to generate a first comparison result, and to calculate the identification coefficient BSX with the preset first threshold value Y to generate a magnitude coefficient LJZ, and to compare the magnitude coefficient LJZ with a preset second threshold value R; The execution unit is used to perform noise reduction according to the second comparison result.

Images