Power dispatch voice command verification method and system based on logic verification

Abstract

The present invention discloses a method and system for verifying electric power dispatching voice instructions based on logic verification, which relates to the technical field of voice recognition and comprises the following steps: obtaining pre-emphasized voice instructions and dividing them into frames; obtaining first instruction data, calculating the impact assessment data of the framing according to the first instruction data and a preset information loss factor, and performing data analysis to obtain the optimal framing length; applying a window function to each frame of voice under the optimal framing length, and performing a fast Fourier transform to obtain the energy distribution and balance data of the voice signal at different frequencies, and adjusting the window function length according to the balance data; performing data analysis on the energy distribution of the voice signal at different frequencies through a Mel filter to obtain a Mel spectrum, and performing a logarithmic conversion to obtain a Mel logarithmic spectrum; performing a cepstrum analysis on the Mel logarithmic spectrum to extract voice features; the present application improves the recognition ability of voice instructions by adjusting the frame length and the length of the window function.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and more specifically, to a method and system for verifying electric power dispatching speech instructions based on logic verification. Background Art

[0002] With the rapid development of artificial intelligence technology, smart devices have gradually penetrated into every aspect of our lives, from smart homes to smart car systems to enterprise-level smart scheduling systems. These devices have greatly improved our quality of life and work efficiency. Among them, speech recognition and speech synthesis technology, as one of the core technologies in smart devices, is playing an increasingly important role.

[0003] Speech recognition technology can convert human voice signals into text information, thus realizing human-computer interaction. This technology has undergone continuous evolution from manually designed rules to statistical methods and then to deep learning. Especially driven by deep learning technology, the performance of speech recognition technology has been significantly improved, and the recognition accuracy has been greatly improved, making it increasingly widely used in smart devices. Speech synthesis technology can convert text information into smooth and natural speech output, providing users with a more vivid and convenient auditory experience.

[0004] In the intelligent device dispatching system, the recognition and verification of voice commands are key links. Traditional dispatching systems often rely on manual input or preset key operations, which is not only cumbersome to operate but also inefficient. The intelligent dispatching system based on voice recognition can quickly and accurately complete the allocation and dispatch of tasks by recognizing the user's voice commands, greatly improving the dispatching efficiency.

[0005] For example, the invention patent with announcement number: CN117636876A discloses a speech verification method and apparatus, a computing device, and a computer-readable storage medium, and the following steps are included: obtaining speech data to be verified, and converting the speech data into text data; based on multiple prompt text groups, calling a target text processing model to process the text data, and obtaining a reply result for each prompt text group, wherein the prompt text group is obtained based on the analysis of the verification rule information in the reference text, and the prompt text group is used to indicate whether the target text processing model outputs a reply result that conforms to the verification rule information; based on the verification weights and reply results of each prompt text group, determining the verification result for the speech data; and reducing the complexity of speech data analysis by the above method.

[0006] For example, the invention patent with announcement number: CN104022879B discloses a method and device for voice security verification, which includes the following steps: when receiving a user's security verification instruction, obtaining the voice data entered by the user; performing a similarity comparison between the voice data and the pre-stored standard voice data according to a preset voice algorithm to obtain the similarity between the voice data and the pre-stored standard voice data; when the similarity is greater than a preset threshold, determining that the user has passed the security verification. The present invention also discloses a device for voice security verification, which effectively reduces the probability of hacker attacks and improves the security of information while ensuring the simplicity of security verification operations.

[0007] The above disclosed technical solutions have at least the following technical problems:

[0008] The conventional power dispatching voice command verification method will greatly increase the probability of misrecognition in a noisy environment and when the user's pronunciation is inaccurate, thereby reducing the accuracy and robustness of voice recognition. In view of the above problems, the present invention proposes a solution. Summary of the invention

[0009] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method and system for verifying electric power dispatching voice commands based on logic verification, which solves the problem of low recognition ability of voice commands by adjusting the frame length and the length of the window function.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] The electric power dispatch voice command verification method based on logic verification includes the following steps: obtaining a pre-emphasized voice command and dividing it into frames, obtaining the first command data, wherein the first command data includes time resolution and frequency resolution, obtaining a data set containing multiple voice characteristics, and performing feature extraction on a frame sequence under each overlapping ratio of the data set, screening the minimum overlapping feature, obtaining an information loss factor, calculating the impact assessment data of the frame division according to the first command data and the preset information loss factor, and performing data analysis on the impact assessment data of the frame division to obtain an optimal frame length; applying a window function to each frame of voice under the optimal frame length, and performing a fast Fourier transform to obtain energy distribution and balance data of the voice signal at different frequencies, and adjusting the window function length according to the balance data; performing data analysis on the energy distribution of the voice signal at different frequencies through a Mel filter to obtain a Mel spectrum, and performing a logarithmic conversion to obtain a Mel logarithmic spectrum; performing a cepstrum analysis on the Mel logarithmic spectrum to extract voice features.

[0012] In a preferred embodiment, the voice command is pre-emphasized as follows: the voice command of the user control device is obtained and a spectrum analysis is performed to filter the high-frequency signal; different pre-emphasis coefficients are adapted to the high-frequency signal based on a filter, and a communication quality test is performed on the high-frequency signal to evaluate the signal index data, wherein the signal index data includes clarity, recognition rate and noise suppression effect; the pre-emphasis coefficient is determined according to the signal index data, and the voice command is pre-emphasized according to the pre-emphasis coefficient.

[0013] In a preferred embodiment, the data analysis to obtain the optimal frame length includes: extracting feature data from the impact assessment data, the feature data including frame length, signal sampling rate and ratio of frame overlap; traversing and calculating the impact assessment data of the frames according to the value range of the feature data to construct an impact assessment data set for the frames; performing comparative analysis based on the data set to obtain the highest impact assessment data for the frames, and obtaining the optimal value of the feature data based on the impact assessment data set for the frames.

[0014] In a preferred embodiment, a window function is applied to each frame of speech under the optimal framing length, and a fast Fourier transform is performed to obtain energy distribution and balance data of the speech signal at different frequencies, and the length of the window function is adjusted according to the balance data, specifically as follows: the speech is divided into multiple groups, and window functions of different lengths are applied to each frame of speech in each group; the windowed frames are fast Fourier transformed, and data analysis is performed to obtain energy distribution and balance data of the speech signal at different frequencies; according to the balance data, the optimal balance coefficient between spectrum leakage and resolution of window functions of different lengths is screened out to obtain the optimal window function length.

[0015] In a preferred embodiment, the steps for acquiring the balanced data are as follows: obtaining the number of frequency components and spectrum components based on the spectrum data obtained by Fourier transform, and calculating the amplitude of the frequency components, screening out the maximum amplitude in the window function spectrum, and calculating the spectrum leakage coefficient; obtaining the width of the main lobe of the window function, the frame length of the speech signal after framing, and the sampling interval of the signal in the time domain, and performing data analysis to obtain the resolution of the window function; inputting the resolution of the window function, the spectrum leakage coefficient and the frame length into a preset model to obtain the balance coefficient between the spectrum leakage and the resolution as the balanced data.

[0016] In a preferred embodiment, the energy distribution of the speech signal at different frequencies is analyzed by a Mel filter to obtain a Mel spectrum, and a logarithmic transformation is performed to obtain a Mel logarithmic spectrum, specifically as follows: the energy of each spectral line is obtained by a Mel filter; the energy of each filter is obtained by inversely deducing the energy of each spectral line, the energy of each filter is used as the corresponding value of the Mel spectrum, and a discrete cosine transform is performed to obtain a Mel spectrum; the Mel spectrum is logarithmically transformed to obtain a Mel logarithmic spectrum.

[0017] In a preferred embodiment, the method of performing cepstrum analysis on the Mel logarithmic spectrum to extract speech features is specifically as follows: performing discrete cosine transform on the Mel logarithmic spectrum to convert the signal from the time domain and frequency domain to the cepstrum domain to obtain cepstrum coefficients; performing differential processing on the processed cepstrum coefficients and performing dynamic feature extraction to obtain speech features to be processed; performing feature selection and dimensionality reduction on the speech features to be processed by principal component analysis and linear discriminant analysis methods to obtain speech features.

[0018] The system of the electric power dispatch voice instruction verification method based on logic verification is characterized in that it includes a framing module, a windowing module, a data conversion module and a feature extraction module, and there is a connection between the modules; the framing module obtains the pre-emphasized voice instruction, and frames it to obtain the first instruction data, the first instruction data includes time resolution and frequency resolution, obtains a data set containing multiple voice characteristics, and extracts features from the frame sequence under each overlapping ratio of the data set, screens the features with the smallest overlap, obtains the information loss factor, calculates the impact evaluation data of the framing according to the first instruction data and the preset information loss factor, and performs the impact evaluation of the framing. The estimated data is analyzed to obtain an optimal frame length, wherein the impact evaluation data of the frame includes a ratio of overlap between frames and the first data; a windowing module is used to apply a window function to each frame of speech under the optimal frame length, and perform a fast Fourier transform to obtain energy distribution and balance data of the speech signal at different frequencies, and adjust the window function length according to the balance data; a data conversion module is used to perform data analysis on the energy distribution of the speech signal at different frequencies through a Mel filter to obtain a Mel spectrum, and perform a logarithmic conversion to obtain a Mel logarithmic spectrum; a feature extraction module is used to perform a cepstrum analysis on the Mel logarithmic spectrum to extract speech features.

[0019] The technical effects and advantages of the power dispatching voice command verification method and system based on logic verification of the present invention are as follows:

[0020] 1. The present invention adopts multi-level signal processing technology, such as pre-emphasis, framing, window function application and fast Fourier transform (FFT), as well as Mel spectrum analysis. These processing methods can effectively reduce the interference of noise and distortion on voice signals and improve the accuracy of voice recognition. Especially in a complex power dispatching environment, factors such as equipment noise and environmental noise may cause recognition errors of voice commands. By finely processing voice signals, the robustness and clarity of voice recognition can be significantly improved.

[0021] 2. The present invention selects the optimal framing scheme according to the actual situation by adjusting the frame length and the overlap ratio between frames. Specifically, the influence of different overlap ratios on the computational overhead is analyzed by linear regression, and the signal processing efficiency is evaluated and optimized by combining the characteristics of time resolution and frequency resolution. This framing optimization can not only reduce the computational burden, but also improve the processing speed, ensure the real-time performance of the system, and meet the demand for rapid response of the power dispatching system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The present invention is a schematic diagram of the structure of a method for verifying electric power dispatching voice commands based on logic verification.

[0023] Figure 2 The present invention is a schematic diagram of the system structure of a method for verifying electric power dispatching voice instructions based on logic verification. DETAILED DESCRIPTION

[0024] 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.

[0025] Embodiment 1, Figure 1 The present invention provides a method for verifying electric power dispatching voice commands based on logic verification, and the specific steps are as follows:

[0026] S1, obtain the pre-emphasized voice instruction, divide it into frames, obtain the first instruction data, the first instruction data includes time resolution and frequency resolution, obtain a data set containing multiple voice characteristics, and perform feature extraction on the frame sequence under each overlapping ratio of the data set, screen the minimum overlapping features, and obtain the information loss factor, calculate the impact evaluation data of the frame division according to the first instruction data and the preset information loss factor, and perform data analysis on the impact evaluation data of the frame division to obtain the optimal frame length, the impact evaluation data of the frame division includes the ratio of inter-frame overlap and the first data.

[0027] In this embodiment, the voice command is pre-emphasized as follows:

[0028] Obtain the user's voice command to control the device, perform spectrum analysis on the input voice signal, and filter out high-frequency signals;

[0029] Adapting different pre-emphasis coefficients to the high-frequency signal based on the filter, and performing communication quality test on the high-frequency signal to evaluate signal index data, wherein the signal index data includes clarity, recognition rate and noise suppression effect;

[0030] A pre-emphasis coefficient is determined according to the signal indicator data, and the voice command is pre-emphasized according to the pre-emphasis coefficient.

[0031] The calculation formula of the filter is as follows:

[0032]

[0033] In the formula, is a pre-emphasized signal. is the input signal, is the pre-emphasis factor, usually between 0.9 and 1.0.

[0034] It should be noted that in voice communication, the frequency range of human voice is usually between 300Hz and 3400Hz, but the high-frequency part (such as above 3kHz) tends to decay faster, and high-frequency attenuation will affect the clarity and intelligibility of the signal.

[0035] Further, the advantages of pre-emphasis are explained:

[0036] Enhance high-frequency information: Through pre-emphasis processing, the proportion of high-frequency components in the speech signal is increased, which helps to better capture high-frequency details in subsequent analysis and processing;

[0037] Avoid numerical problems in FFT operations: The enhancement of high-frequency components can reduce numerical instability and errors in FFT operations and improve the accuracy of analysis;

[0038] Increase signal-to-noise ratio: Pre-emphasis can enhance the strength of useful signals (especially high-frequency parts), making useful signals easier to identify and extract compared to background noise. This helps improve the accuracy and robustness of speech recognition;

[0039] Improve distortion caused by uneven frequency response: During transmission, high-frequency signals are usually attenuated due to limitations of lines and equipment. Pre-emphasis can increase high-frequency energy at the source to offset the loss in the subsequent attenuation process, thereby reducing distortion.

[0040] In this embodiment, the impact assessment data of the sub-frame is calculated according to the instruction first data and the preset information loss factor, and the impact assessment data of the sub-frame is analyzed to obtain the optimal sub-frame length, which is as follows:

[0041] Divide the pre-emphasized speech signal into frames of different lengths to obtain impact evaluation data of the frame division;

[0042] The steps for obtaining the impact assessment data of the frame are as follows:

[0043] Obtain the pre-emphasized voice command, divide the voice command into frames, divide it into multiple groups of frames with different lengths, and record the length of each group of frames;

[0044] The signal sampling rate when obtaining the voice command, and the time resolution and frequency resolution are obtained according to the ratio of the frame length to the signal sampling rate;

[0045] Obtain a data set containing multiple speech features, and adjust the inter-frame overlap ratio from 0 to 1 according to a fixed step size during the framing process;

[0046] Perform feature extraction on the frame sequence at each overlapping ratio, and select the feature with the minimum overlap as the benchmark;

[0047] By comparing the data between the corresponding frame sequence features and the reference features, the information loss factor is obtained;

[0048] The information loss factor is an indicator to measure the degree of feature information loss caused by the adjustment of the overlap ratio between frames during the framing process. The information loss is quantified by comparing the data difference between the frame sequence features and the reference features under different overlap ratios.

[0049] Obtain the computational costs under different overlap ratios between frames to obtain a cost data set;

[0050] Linear regression analysis is performed on the overhead data set to obtain the additional calculation factor caused by the overlap between frames;

[0051] The processing efficiency is obtained through data analysis based on the additional calculation factors caused by the overlap between frames, the frame length and the signal sampling rate;

[0052] Input the time resolution, frequency resolution, information loss factor and processing efficiency into a preset frame impact assessment formula, and output the frame impact assessment data;

[0053] Extracting feature data from the impact assessment data, the feature data including frame length, signal sampling rate, and ratio of inter-frame overlap;

[0054] The impact assessment data of the sub-frames are traversed and calculated according to the value range of the feature data to construct an impact assessment data set for the sub-frames;

[0055] The highest sub-frame impact assessment data is obtained by performing comparative analysis on the data set, and the optimal value of the feature data is obtained based on the sub-frame impact assessment data set.

[0056] The calculation formulas for the time resolution and frequency resolution are as follows:

[0057]

[0058] The calculation formula of the information loss factor is as follows:

[0059]

[0060] The calculation formula of the processing efficiency is as follows:

[0061]

[0062] The calculation formula of the impact assessment data of the frame is as follows:

[0063]

[0064] In the formula, is the temporal resolution, is the frequency resolution, is the frame length, is the signal sampling rate (samples / second), is the information loss factor, is a constant factor, is the ratio of frame overlap, is the processing efficiency of framing, is the additional calculation factor caused by the overlap between frames, and c>0 is a constant. represents the correction factor for the computational overhead of overlap, It is the impact assessment data of the frame.

[0065] It should be noted that a larger constant factor means that for a larger overlap ratio, information loss decreases faster; the higher the sampling rate, the higher the time resolution of the signal and the lower the frequency resolution; the higher the impact assessment data of the frame, the stronger the overall impact, which may mean that the response or effect of the system is more significant; the lower the impact assessment data of the frame, the weaker the overall impact and the weaker the response or effect of the system.

[0066] S2, apply the window function to each frame of speech under the optimal frame length, and perform fast Fourier transform to obtain the energy distribution and balance data of the speech signal at different frequencies, and adjust the window function length according to the balance data.

[0067] In this embodiment, the window function is applied to each frame of speech under the optimal framing length, and a fast Fourier transform is performed to obtain the energy distribution and balance data of the speech signal at different frequencies, and the window function length is adjusted according to the balance data, as follows:

[0068] Divide the speech into multiple groups, apply window functions of different lengths to each frame of speech in each group, and reduce spectral leakage by smoothing the discontinuity of the window edge;

[0069] Perform fast Fourier transform on the windowed frames and perform data analysis to obtain the energy distribution and balance data of the speech signal at different frequencies;

[0070] According to the balanced data, the optimal balance coefficient between the spectrum leakage and the resolution of the window function with different lengths is screened out to obtain the optimal window function length.

[0071] The steps for obtaining the balance data are as follows:

[0072] Perform window function processing on the speech signal and perform Fourier transform to obtain spectrum data;

[0073] Obtain the number of frequency components and spectrum components according to the spectrum data obtained by Fourier transform, calculate the amplitude of the frequency component, filter out the maximum amplitude in the window function spectrum, and calculate the spectrum leakage coefficient;

[0074] Obtain the type of window function, and obtain the width of the main lobe of the window function according to the frequency spectrum of the window function;

[0075] Obtain the frame length of the speech signal after framing and the sampling interval of the signal in the time domain, and perform data analysis in combination with the width of the main lobe of the window function to obtain the resolution of the window function;

[0076] The resolution of the window function, the spectrum leakage coefficient and the frame length are input into a preset balance calculation formula to obtain a balance coefficient between the spectrum leakage and the resolution as balance data.

[0077] The calculation formula of the spectrum leakage is as follows:

[0078]

[0079] The calculation formula of the resolution is as follows:

[0080]

[0081] The calculation formula of the balance data is as follows:

[0082]

[0083] In the formula, is the spectrum leakage coefficient, is each frequency component in the spectrum of the window function, represents the maximum amplitude in the window function spectrum, is the number of components in the window function spectrum, is the resolution factor, is the width of the main lobe of the window function, is the frame length, is the frequency response of the window function, is the sampling period, It is balanced data.

[0084] It should be noted that the widespread use of various nonlinear loads in power systems will cause grid waveform distortion and generate harmonic interference. Harmonic interference will make the signal spectrum complex and increase the risk of spectrum leakage.

[0085] To further explain, spectrum leakage refers to the phenomenon that the signal energy is extended to other frequency components in the frequency domain, especially when the signal frequency components are not completely located at discrete frequency points. Spectral leakage is usually caused by the shape of the window function. Different window function shapes have different leakage characteristics. The size of spectrum leakage is related to the main lobe width and side lobe attenuation of the window function. For example, a rectangular window function will produce a large spectrum leakage because its main lobe is wide and the side lobe decays slowly. Window functions such as Gaussian window, Hamming window, and Heyman window can reduce spectrum leakage; resolution is directly related to the length and shape of the window function. Generally speaking, the longer the frame length (that is, the longer the length of the window function), the higher the resolution of the spectrum, and the more accurately the different frequency components in the signal can be separated. However, an overly long window function will lead to an increase in spectrum leakage.

[0086] S3, performing data analysis on the energy distribution of the speech signal at different frequencies through a Mel filter to obtain a Mel spectrum, and performing logarithmic conversion to obtain a Mel logarithmic spectrum.

[0087] In this embodiment, the Mel filter performs data analysis on the energy distribution of the speech signal at different frequencies to obtain a Mel spectrum, and performs logarithmic conversion to obtain a Mel logarithmic spectrum, as follows:

[0088] Design a Mel filter bank, use the Mel filter to square the Fourier transform result to get the energy of each spectral line;

[0089] The energy of the spectral line calculated above is passed through the designed Mel filter bank, the energy in each filter is calculated, and the energy in each filter is used as the corresponding value of the Mel spectrum, thereby obtaining a complete Mel spectrum;

[0090] Performing discrete cosine transform on the Mel spectrum further reduces the dimension of the Mel spectrum and removes correlation;

[0091] Apply the logarithmic function to each frequency component of the Mel spectrum and obtain the result of the logarithmic operation, that is, the Mel logarithmic spectrum.

[0092] The specific steps of designing the Mel filter bank are as follows:

[0093] Determine the number of Mel filters based on the real-time performance of the real-time speech signal and the resource limitations of the hardware platform;

[0094] Get the frequency range of the filter, that is, the minimum frequency and maximum frequency, the minimum frequency is 0, and the maximum frequency is half of the sampling frequency;

[0095] Convert the filter frequency range to Mel scale by using the conversion formula between Mel scale and Hertz;

[0096] On the converted Mel scale, take the determined number of filter points evenly to obtain a series of Mel scale values, convert the Mel scale values ​​back to Hertz to obtain the center frequency of the filter;

[0097] According to the center frequency obtained above, the shape of the filter is designed, and the frequency response of the filter is 1 at the center frequency, and then linearly decreases to 0 at the center frequency of the adjacent filter.

[0098] S4, performs cepstrum analysis on the Mel log spectrum to extract speech features.

[0099] In this embodiment, the cepstrum analysis is performed on the Mel logarithmic spectrum to extract speech features, as follows:

[0100] Perform discrete cosine transform on the Mel logarithmic spectrum, convert the signal from the time domain and frequency domain to the cepstrum domain, and perform normalization;

[0101] Perform differential processing on the processed cepstral coefficients, perform dynamic feature extraction, and obtain first-order differential and second-order differential coefficients;

[0102] Principal component analysis and linear discriminant analysis methods are used to perform feature selection and dimensionality reduction to reduce redundant information.

[0103] Embodiment 2, Figure 2 A system for verifying a power dispatching voice command based on logic verification is provided, which is characterized by comprising a framing module, a windowing module, a data conversion module and a feature extraction module, and there are connections between the modules;

[0104] A framing module is used to obtain the pre-emphasized voice instruction, and to perform framing to obtain the first instruction data, wherein the first instruction data includes time resolution and frequency resolution, obtain a data set containing multiple voice characteristics, and perform feature extraction on the frame sequence under each overlapping ratio of the data set, select the feature with the minimum overlap, calculate the impact evaluation data of the framing according to the first instruction data and a preset information loss factor, and perform data analysis on the impact evaluation data of the framing to obtain the optimal framing length;

[0105] The windowing module is used to apply the window function to each frame of speech under the optimal frame length, and perform fast Fourier transform to obtain the energy distribution and balance data of the speech signal at different frequencies, and adjust the window function length according to the balance data;

[0106] A data conversion module is used to perform data analysis on the energy distribution of the speech signal at different frequencies through a Mel filter to obtain a Mel spectrum, and perform logarithmic conversion to obtain a Mel logarithmic spectrum;

[0107] The feature extraction module is used to perform cepstrum analysis on the Mel log spectrum and extract speech features.

[0108] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0109] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0110] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0111] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0112] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0113] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A power dispatch voice command verification method based on logic verification, characterized in that: The steps include: Get the pre-emphasized voice command and divide it into frames; Acquire first instruction data, wherein the first instruction data includes a time resolution and a frequency resolution; Obtain a data set containing multiple speech characteristics, perform feature extraction on the frame sequence at each overlapping ratio of the data set, select the feature with the minimum overlap, and obtain the information loss factor; Calculating the impact assessment data of the sub-frames according to the first instruction data and the preset information loss factor, and performing data analysis on the impact assessment data of the sub-frames to obtain the optimal sub-frame length, wherein the impact assessment data of the sub-frames includes the ratio of inter-frame overlap and the first data; Apply the window function to each frame of speech under the optimal frame length, and perform fast Fourier transform to obtain the energy distribution and balance data of the speech signal at different frequencies, and adjust the window function length according to the balance data; The energy distribution of the speech signal at different frequencies is analyzed by the Mel filter to obtain the Mel spectrum, and then the Mel logarithmic spectrum is obtained by logarithmic conversion. Perform cepstrum analysis on the Mel log spectrum to extract speech features; The calculation formula of the information loss factor is as follows: The calculation formula of the impact assessment data of the frame is as follows: In the formula, is the temporal resolution, is the frequency resolution, is the frame length, is the information loss factor, is the ratio of frame overlap, is the processing efficiency of framing, is the frame-by-frame impact assessment data, is a constant factor, It is the natural base.

2. The method for verifying electric power dispatching voice instructions based on logic verification according to claim 1, characterized in that: Pre-emphasize the voice commands as follows: Obtain the user's voice commands to control the device and perform spectrum analysis to filter high-frequency signals; Adapting different pre-emphasis coefficients to the high-frequency signal based on the filter, and performing communication quality test on the high-frequency signal to evaluate signal index data, wherein the signal index data includes clarity, recognition rate and noise suppression effect; A pre-emphasis coefficient is determined according to the signal indicator data, and the voice command is pre-emphasized according to the pre-emphasis coefficient.

3. The method for verifying electric power dispatching voice instructions based on logic verification according to claim 2 is characterized in that: The data analysis is performed to obtain the optimal frame length, which is specifically as follows: Extracting feature data from the framed impact assessment data, the feature data including frame length, signal sampling rate, and ratio of inter-frame overlap; The impact assessment data of the sub-frames are traversed and calculated according to the value range of the feature data to construct an impact assessment data set of the sub-frames; The highest sub-frame impact assessment data is obtained by performing comparative analysis on the data set, and the optimal value of the feature data is obtained based on the sub-frame impact assessment data set.

4. The method for verifying electric power dispatching voice instructions based on logic verification according to claim 3 is characterized in that: The window function is applied to each frame of speech under the optimal framing length, and a fast Fourier transform is performed to obtain the energy distribution and balance data of the speech signal at different frequencies, and the window function length is adjusted according to the balance data, specifically: Divide the speech into multiple groups, and apply window functions of different lengths to each frame of speech in each group; Perform fast Fourier transform on the windowed frames and perform data analysis to obtain the energy distribution and balance data of the speech signal at different frequencies; According to the balanced data, the optimal balance coefficient between the spectrum leakage and the resolution of the window function under different lengths is screened out to obtain the optimal window function length.

5. The method for verifying electric power dispatching voice instructions based on logic verification according to claim 4 is characterized in that: The steps for obtaining the balance data are as follows: Obtain the number of frequency components and spectrum components according to the spectrum data obtained by Fourier transform, calculate the amplitude of the frequency component, filter out the maximum amplitude in the window function spectrum, and calculate the spectrum leakage coefficient; Obtain the width of the main lobe of the window function, the frame length of the speech signal after framing, and the sampling interval of the signal in the time domain, and perform data analysis to obtain the resolution of the window function; The resolution, spectrum leakage coefficient and frame length of the window function are input into a preset model to obtain a balance coefficient between spectrum leakage and resolution as balance data.

6. The method for verifying electric power dispatching voice instructions based on logic verification according to claim 5 is characterized in that: The energy distribution of the speech signal at different frequencies is analyzed by the Mel filter to obtain the Mel spectrum, and the Mel logarithmic spectrum is obtained by logarithmic conversion, as follows: Get the energy of each spectral line through the Mel filter; The energy of each filter is obtained by inverse calculation based on the energy of each spectral line. The energy of each filter is used as the corresponding value of the Mel spectrum, and a discrete cosine transform is performed to obtain the Mel spectrum. Perform logarithmic transformation on the Mel spectrum to obtain the Mel log spectrum.

7. The method for verifying electric power dispatching voice instructions based on logic verification according to claim 6, characterized in that: The cepstrum analysis is performed on the Mel logarithmic spectrum to extract speech features, specifically as follows: Perform discrete cosine transform on the Mel logarithmic spectrum to convert the signal from the time domain and frequency domain to the cepstrum domain to obtain the cepstrum coefficients; Performing differential processing on the processed cepstral coefficients and performing dynamic feature extraction to obtain the speech features to be processed; The principal component analysis and linear discriminant analysis methods are used to perform feature selection and dimension reduction on the speech features to be processed to obtain speech features.

8. A system using the power dispatch voice command verification method based on logic verification as claimed in any one of claims 1 to 7, characterized in that: It includes a framing module, a windowing module, a data conversion module and a feature extraction module, and there are connections between the modules; A framing module is used to obtain the pre-emphasized voice instruction, and to perform framing to obtain first instruction data, wherein the first instruction data includes time resolution and frequency resolution, obtain a data set containing multiple voice characteristics, and perform feature extraction on a frame sequence at each overlapping ratio of the data set, screen the feature with the minimum overlap, obtain an information loss factor, calculate the impact assessment data of the framing according to the first instruction data and the preset information loss factor, and perform data analysis on the impact assessment data of the framing to obtain an optimal framing length, wherein the impact assessment data of the framing includes the ratio of inter-frame overlap and the first data; The windowing module is used to apply the window function to each frame of speech under the optimal frame length, and perform fast Fourier transform to obtain the energy distribution and balance data of the speech signal at different frequencies, and adjust the window function length according to the balance data; A data conversion module is used to perform data analysis on the energy distribution of the speech signal at different frequencies through a Mel filter to obtain a Mel spectrum, and perform logarithmic conversion to obtain a Mel logarithmic spectrum; The feature extraction module is used to perform cepstrum analysis on the Mel log spectrum and extract speech features.

Images