Cross-domain voice recognition model optimization method

Abstract

The invention belongs to the technical field of intelligent sound recognition, and provides a cross-domain sound recognition model optimization method, which comprises the following steps: in a recognition capture period, importing different domain sound features into a cross-domain sound recognition model to obtain a plurality of groups of time-frequency feature sequences, and performing feature difference analysis according to the plurality of groups of time-frequency feature sequences to obtain a feature difference value; and the change characteristic difference of the time-frequency characteristics of the sounds in different domains in the time domain and the intensity difference degree of the time-frequency characteristics of the sounds in different domains in the frequency domain are reflected through the characteristic difference value, so that the recognition of the different sounds is completed, the difference characteristics of the different sounds are determined, and the recognition accuracy of the cross-domain sound recognition model is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent sound recognition, and in particular to a cross-domain sound recognition model optimization method. Background Art

[0002] In the field of sound recognition, with the continuous expansion of application scenarios, the sources and types of sounds that need to be recognized are becoming increasingly complex and diverse, and cross-domain sound recognition faces many challenges. Traditional sound recognition models often have the problem of low recognition accuracy when processing sound data from different domains. At the same time, in terms of model storage and performance optimization, they are unable to effectively handle the storage problem of a large number of differential features, resulting in excessive storage burden on the model, affecting its operating efficiency and recognition accuracy.

[0003] In the prior art, the model cannot effectively handle the storage problem of a large number of difference features, resulting in an excessive storage burden on the model, affecting its operating efficiency and recognition accuracy. Therefore, the present application reflects the difference in the change characteristics of the time-frequency features of sounds in different domains in the time domain and the degree of difference in the intensity of the time-frequency features of sounds in different domains in the frequency domain through feature difference values, thereby completing the recognition of different sounds, determining the difference features of different sounds, and improving the recognition accuracy of the cross-domain sound recognition model. The storage ratio of the difference features of the cross-domain sound recognition model in multiple historical storage cycles is reflected through the storage ratio value, which is conducive to evaluating the storage ratio of the difference features of the cross-domain sound recognition model. At the same time, the priority storage coefficient is obtained, and the difference features are preferentially stored according to the priority storage coefficient. When the priority storage memory is equal to the storage ratio threshold, the difference feature storage work is stopped, thereby solving the problem that the cross-domain sound recognition model cannot be completely stored when storing multiple difference features due to the large memory occupancy of multiple difference features, and then reasonably optimizing the storage to reduce the storage burden of the cross-domain sound recognition model and improve the recognition accuracy of the cross-domain sound recognition model for sounds in different domains.

[0004] To this end, the present invention provides a cross-domain sound recognition model optimization method. Summary of the invention

[0005] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0006] The technical solution adopted by the present invention to solve its technical problem is: The cross-domain sound recognition model optimization method includes the following modules: Step 1: In the recognition and capture cycle, the features of sounds in different domains are imported into the cross-domain sound recognition model, and the time-frequency features are extracted to obtain multiple groups of time-frequency feature sequences; Step 2: Perform feature difference analysis on the time-frequency feature sequences corresponding to sounds in different domains to obtain zero rate difference values ​​and formant difference values, perform quantization processing to obtain difference features, and summarize to obtain a feature difference degree sequence; Step 3: In multiple historical storage cycles, obtain the feature byte ratio and feature quantity ratio, perform quantification processing, and compare them with the memory corresponding to all difference features in the feature difference degree sequence to generate a storage optimization signal; Step 4: Based on the storage optimization signal, obtain the priority storage coefficient, and complete the priority storage of the difference features according to the priority storage coefficient.

[0007] As a further solution of the present invention, the extraction process of time-frequency features is as follows: Divide the recognition and capture cycle into a number of capture time periods at equal time intervals; The time-frequency features are processed by short-time Wavelet transform to obtain the time-frequency feature values; The time-frequency feature values ​​corresponding to all capture time periods in the recognition capture period of different sounds are sorted according to the time series corresponding to the capture time periods to obtain multiple groups of time-frequency feature sequences.

[0008] As a further solution of the present invention, the process of obtaining the zero rate difference value is as follows: A time-frequency feature sequence is randomly extracted as a feature comparison sequence, and the remaining time-frequency feature sequences after the extraction are combined with the feature comparison sequence one by one to form a comparison analysis group, thereby obtaining multiple comparison analysis groups; Randomly select a comparison analysis group, substitute the time-frequency feature values ​​corresponding to all time-frequency features into the zero-crossing rate formula, and obtain the zero-crossing rate corresponding to the time-frequency feature of each capture node; By obtaining the zero-crossing rates corresponding to all time-frequency features in the feature comparison sequence, the zero-crossing rates corresponding to all time-frequency features in the remaining time-frequency feature sequence after extraction are obtained, and the zero-crossing rate difference value is obtained through the formula.

[0009] As a further solution of the present invention, the resonance peak difference value is obtained by: Construct a feature comparison change curve, extract the time-frequency feature values ​​corresponding to all peaks, and integrate them according to the time series to obtain a comparison peak set; Construct a time-frequency feature change curve, extract the time-frequency feature values ​​corresponding to all peaks, and integrate them according to the time series to obtain a set of time-frequency feature peaks; Substitute the comparison peak set and the time-frequency feature peak set into the Euclidean distance formula to obtain the resonance peak difference value ; Extract the time-frequency feature values ​​corresponding to all troughs and integrate them according to the time series to obtain the comparison trough set; Extract the time-frequency feature values ​​corresponding to all troughs and integrate them according to the time series to obtain the time-frequency feature trough set; Substitute the comparison trough set and the time-frequency feature trough set into the Euclidean distance formula to obtain the resonance trough difference value ; The resonance peak difference Difference from resonance trough Perform sum calculation to obtain the resonance peak difference value.

[0010] As a further solution of the present invention, the characteristic difference degree sequence is obtained as follows: The zero-crossing rate difference value and the resonance peak difference value are summed up to obtain a characteristic difference value; If the feature difference value is greater than the feature difference threshold, a difference feature signal is generated, and the time-frequency feature of the generated difference feature signal is marked as a difference feature; All difference features are sorted from large to small according to the corresponding feature difference values ​​to obtain a feature difference degree sequence.

[0011] As a further solution of the present invention, the characteristic quantity ratio is obtained as follows: Arbitrarily extract a historical storage period; Divide the historical storage cycle into several historical storage periods, obtain the number of stored difference features in the historical storage period, and calculate the ratio with the number of all features corresponding to the sound to obtain the unit storage quantity ratio; The unit storage quantity ratios corresponding to all historical storage periods in the historical storage cycle are summed and averaged to obtain the period storage quantity ratio; The period storage quantity ratios corresponding to all historical storage periods are compared, the maximum period storage quantity ratio and the minimum period storage quantity ratio are selected, and the sum and mean are calculated to obtain the characteristic quantity ratio.

[0012] As a further solution of the present invention, the characteristic byte ratio is obtained as follows: Obtain the bytes corresponding to the storage difference features during the historical storage period, and calculate the ratio with the total bytes of all features corresponding to the sound to obtain the unit storage byte ratio; The unit storage byte ratios corresponding to all historical storage periods in the historical storage cycle are summed and averaged to obtain the period storage byte ratio; The cycle storage byte ratios corresponding to all historical storage cycles are compared, the maximum cycle storage byte ratio and the minimum cycle storage byte ratio are selected, and the summed average is calculated to obtain the characteristic byte ratio.

[0013] As a further solution of the present invention, the generation process of the storage optimization signal is as follows: The feature byte ratio is multiplied by the feature quantity ratio to obtain the storage ratio value; If the storage ratio value is less than or equal to the storage ratio threshold, a storage optimization signal is generated.

[0014] As a further solution of the present invention, the priority storage coefficient is obtained in the following manner: Extract the bytes corresponding to the difference features, calculate the ratio with the total amount of bytes corresponding to all the difference features, obtain the feature byte value, and sort all the difference features from large to small according to the feature byte values ​​corresponding to the difference features, and obtain the feature byte storage sequence; Obtain the byte of any difference feature in the feature difference degree sequence, calculate the ratio with the total bytes of all features corresponding to the sound, and perform summation and mean calculation to obtain the feature difference byte ratio; The feature byte storage sequence is overlapped and paired with the feature difference degree sequence, the difference features of the overlapped pairing are extracted, and the corresponding feature difference value and feature difference byte ratio are obtained, and the feature difference value and feature difference byte ratio are calculated to obtain the priority storage coefficient.

[0015] As a further solution of the present invention, the difference features are preferentially stored according to the priority storage coefficient, and the process is as follows: Compare the priority storage coefficients corresponding to the difference features in the feature difference degree sequence, and store the difference features in a priority order from large to small; The feature difference byte ratios corresponding to the priority stored difference features are dynamically counted to obtain the priority storage memory, until the priority storage memory is equal to the storage ratio threshold, and the difference feature storage work is stopped.

[0016] The beneficial effects of the present invention are as follows: The present invention imports different domain sound features into a cross-domain sound recognition model within a recognition and capture cycle to obtain multiple groups of time-frequency feature sequences, and performs feature difference analysis based on the multiple groups of time-frequency feature sequences to obtain feature difference values. The feature difference values ​​reflect the difference in the change characteristics of the time-frequency features of sounds in different domains in the time domain, as well as the degree of difference in the intensity of the time-frequency features of sounds in different domains in the frequency domain, thereby completing the recognition of different sounds, determining the difference features of different sounds, and improving the recognition accuracy of the cross-domain sound recognition model. Based on the difference feature signal, the present invention obtains the memory information of the difference features in multiple historical storage cycles through PyTorch, and performs storage ratio analysis to obtain the storage ratio value. The storage ratio value reflects the storage ratio of the difference features by the cross-domain sound recognition model in multiple historical storage cycles, which is conducive to evaluating the storage content ratio of the difference features by the cross-domain sound recognition model, thereby storing the difference features of different sounds in a targeted manner, and providing data support for improving the accuracy of sound recognition by the cross-domain sound recognition model; The present invention obtains a priority storage coefficient based on a storage optimization signal, and prioritizes the storage of difference features according to the priority storage coefficient. At the same time, the priority storage memory is dynamically obtained. When the priority storage memory is equal to the storage ratio threshold, the difference feature storage work is stopped, thereby solving the problem that when a cross-domain sound recognition model stores multiple difference features, the multiple difference features occupy a large memory and cannot be completely stored. The storage is then reasonably optimized to reduce the storage burden of the cross-domain sound recognition model, and the recognition accuracy of the cross-domain sound recognition model for sounds in different domains is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the accompanying drawings.

[0018] Figure 1 It is a flow chart of the cross-domain sound recognition model optimization method of the present invention. DETAILED DESCRIPTION

[0019] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0020] Example 1 like Figure 1 As shown, the cross-domain sound recognition model optimization method described in the embodiment of the present invention includes: Step 1: In the recognition and capture cycle, the features of sounds in different domains are imported into the cross-domain sound recognition model to extract time-frequency features, and multiple groups of time-frequency feature sequences are obtained; In some embodiments, the identification capture period is divided into a number of capture time periods at equal time intervals; The time-frequency features of the sound are extracted through short-time Wavelet transform. The process is as follows: A1, select window function ; It should be noted that the window function The role of is to only consider the window function when analyzing the time-frequency characteristics of the sound The characteristics of the sound during the capture period covered; A2, divide the capture period into n capture nodes with equal time intervals, and the time intervals between adjacent capture nodes are equal; The sound feature corresponding to each capture node is marked as , where T is the frame shift, i.e. the time interval between adjacent capture nodes, Represented as the original characteristic value of the sound; A3, the sound features corresponding to each capture node Perform discrete Fourier transform to transform it from time domain to frequency domain. The formula is: , calculate the time-frequency eigenvalue corresponding to the nth capture node ,in, =0, 1, ..., n-1, is an imaginary number; The time-frequency feature values ​​corresponding to all capture periods of the time-frequency features of different sounds within the recognition and capture period are sorted according to the time series corresponding to the capture period to obtain multiple groups of time-frequency feature sequences; Step 2: Perform feature difference analysis on the time-frequency feature sequences corresponding to sounds in different domains to obtain feature difference data, wherein the feature difference data includes zero-crossing rate difference values ​​and formant difference values, quantize the zero-crossing rate difference values ​​and the formant difference values ​​to obtain feature difference values, compare them with feature difference thresholds to obtain difference features, summarize the difference features to obtain a feature difference degree sequence; In some embodiments, a time-frequency feature sequence is arbitrarily extracted as a feature comparison sequence; The remaining time-frequency feature sequences after extraction are combined with the feature comparison sequences one by one to form a comparison analysis group, thereby obtaining multiple comparison analysis groups; Exemplarily, a comparison analysis group is randomly selected; Substitute the time-frequency feature values ​​corresponding to all time-frequency features in the feature comparison sequence into the formula:

[0021] Calculate the zero-crossing rate corresponding to all time-frequency features in the feature comparison sequence , where T represents the time interval between adjacent capture nodes, For the The time-frequency feature values ​​corresponding to the time-frequency features of the capture nodes, It is represented by the time-frequency feature value corresponding to the time-frequency feature of the nth capture node. (·) is the coincidence function; When X>0, (X) = 1, when X = 0, (X) = 0, when X < 0, (X) = -1; It should be noted that the zero-crossing rate corresponding to all time-frequency features in the feature comparison sequence is calculated by counting the number of times that adjacent capture nodes meet the change , divided by is to normalize the zero-crossing rate to the number of zero-crossings per unit time; similarly, the zero-crossing rates corresponding to all time-frequency features in the acquired feature comparison sequence are The zero-crossing rate corresponding to all time-frequency features in the remaining time-frequency feature sequence after extraction is obtained by ; The difference between the zero-crossing rates within the comparison analysis group is calculated by the formula to obtain the zero-crossing rate difference value. The process is as follows: A1, align the time-frequency features of the two sounds in the comparison analysis group in time and frequency to ensure the same time and frequency, as well as the time interval between the same adjacent capture nodes; A2, by formula: , calculate the zero-crossing rate difference ; Exemplarily, in the comparison analysis group, all time-frequency features in the feature comparison sequence are substituted into a two-dimensional coordinate system to obtain a feature comparison change curve, wherein the X-axis is time and the Y-axis is the time-frequency feature; Extract the time-frequency characteristic values ​​corresponding to all peaks in the feature comparison change curve, and integrate them according to the time series to obtain the comparison peak set ,in, It is represented as the time-frequency feature value corresponding to the mth peak in the feature comparison change curve, and m is represented as the total number of time-frequency feature values ​​corresponding to the peaks in the feature comparison change curve; Similarly, all the time-frequency features in the remaining time-frequency feature sequence after extraction are substituted into the two-dimensional coordinate system to obtain the time-frequency feature change curve, where the X-axis is time and the Y-axis is the time-frequency feature; Extract the time-frequency characteristic values ​​corresponding to all the peaks in the time-frequency characteristic change curve, and integrate them according to the time series to obtain the time-frequency characteristic peak set ,in, It is represented by the time-frequency characteristic value corresponding to the mth peak in the time-frequency characteristic change curve, and m is represented by the total number of time-frequency characteristic values ​​corresponding to the peaks in the time-frequency characteristic change curve; It should be noted that the total number of time-frequency feature values ​​corresponding to the peaks in the time-frequency feature change curve is the same as the total number of time-frequency feature values ​​corresponding to the peaks in the feature comparison change curve; By the Euclidean distance formula: , calculate the resonance peak difference value ,in, It is expressed as the time-frequency feature value corresponding to the mth peak in the feature comparison change curve. It is represented by the time-frequency feature value corresponding to the mth peak in the time-frequency feature change curve, and m is represented by the total number of time-frequency feature values ​​corresponding to the peaks in the feature comparison change curve, and the total number of time-frequency feature values ​​corresponding to the peaks in the time-frequency feature change curve; Similarly, the time-frequency feature values ​​corresponding to all the troughs in the feature comparison change curve are extracted and integrated according to the time series to obtain the comparison trough set ,in, It is represented as the time-frequency feature value corresponding to the rth trough in the feature comparison change curve, and r is represented as the total number of time-frequency feature values ​​corresponding to the trough in the feature comparison change curve; Extract the time-frequency characteristic values ​​corresponding to all troughs in the time-frequency characteristic change curve, and integrate them according to the time series to obtain the time-frequency characteristic trough set ,in, It is represented by the time-frequency characteristic value corresponding to the r-th peak in the time-frequency characteristic change curve, and r is represented by the total number of time-frequency characteristic values ​​corresponding to the peak in the time-frequency characteristic change curve; It should be noted that the total number of time-frequency feature values ​​corresponding to the troughs in the time-frequency feature change curve is the same as the total number of time-frequency feature values ​​corresponding to the troughs in the feature comparison change curve; By the Euclidean distance formula: , calculate the resonance trough difference value ,in, It is expressed as the time-frequency characteristic value corresponding to the rth peak in the feature comparison change curve, It is expressed as the time-frequency feature value corresponding to the rth peak in the time-frequency feature change curve, and r is expressed as the total number of time-frequency feature values ​​corresponding to the peaks in the feature comparison change curve, and the total number of time-frequency feature values ​​corresponding to the peaks in the time-frequency feature change curve; The resonance peak difference Difference from resonance trough Perform sum calculation to obtain the resonance peak difference value; The zero crossing rate difference The sum of the difference value of the resonance peak is calculated to obtain the characteristic difference value; It can be understood that the meaning of the characteristic difference value is: it combines the characteristic differences of the sound in the time domain (zero crossing rate) and the frequency domain (resonance peak), and is used to measure the overall difference between the time-frequency characteristic sequences of sounds in different domains. On the one hand, the zero crossing rate difference value reflects the difference in the number of times the signal crosses the zero value per unit time of the time-frequency characteristic sequences of sounds in different domains, and reflects the difference in the change characteristics of the time-frequency characteristics of sounds in different domains in the time domain. On the other hand, the resonance peak difference value reflects the difference in the intensity of the time-frequency characteristics of sounds in different domains in the frequency domain. The feature difference value is compared with the feature difference threshold, and the process is as follows: If the feature difference value is greater than the feature difference threshold, it means that the time-frequency features of the sounds in different domains have large differences in the change characteristics in the time domain and a large difference in the intensity in the frequency domain, and a difference feature signal is generated, and the time-frequency feature that generates the difference feature signal is marked as a difference feature; Sort all difference features from large to small according to the corresponding feature difference values ​​to obtain a feature difference degree sequence; If the feature difference value is less than or equal to the feature difference threshold, it means that the time-frequency features of the sounds in different domains have small differences in the change characteristics in the time domain and small differences in the intensity in the frequency domain, and a non-difference feature signal is generated, and the time-frequency feature that generates the non-difference feature signal is marked as a non-difference feature; The specific implementation method of the embodiment of the present invention is as follows: within the recognition and capture cycle, sound features of different domains are imported into the cross-domain sound recognition model to obtain multiple groups of time-frequency feature sequences, and feature difference analysis is performed based on the multiple groups of time-frequency feature sequences to obtain feature difference values. The feature difference values ​​reflect the differences in the changing characteristics of the time-frequency features of sounds in different domains in the time domain, as well as the degree of difference in the intensity of the time-frequency features of sounds in different domains in the frequency domain, thereby completing the recognition of different sounds, determining the difference features of different sounds, and improving the recognition accuracy of the cross-domain sound recognition model.

[0022] Example 2 like Figure 1 As shown, based on Example 1, the cross-domain sound recognition model optimization method described in the embodiment of the present invention includes: Step 3: In multiple historical storage cycles, the memory information of the difference features is obtained through PyTorch, where the memory information includes the feature byte ratio and the feature quantity ratio. The feature byte ratio and the feature quantity ratio are quantified to obtain the storage ratio value, and compared with the storage ratio threshold. If the storage ratio value is greater than the storage ratio threshold, a storage optimization signal is generated; In some embodiments, one historical storage period is randomly selected from among a plurality of historical storage periods; Divide the historical storage cycle into several historical storage periods, obtain the number of stored difference features in the historical storage period, and calculate the ratio with the number of all features corresponding to the sound to obtain the unit storage quantity ratio; The unit storage quantity ratios corresponding to all historical storage periods in the historical storage cycle are summed and averaged to obtain the period storage quantity ratio; Compare the period storage quantity ratios corresponding to all historical storage periods, select the maximum period storage quantity ratio and the minimum period storage quantity ratio, and calculate the sum and mean to obtain the characteristic quantity ratio; Obtain the bytes corresponding to the storage difference features during the historical storage period, and calculate the ratio with the total bytes of all features corresponding to the sound to obtain the unit storage byte ratio; The unit storage byte ratios corresponding to all historical storage periods in the historical storage cycle are summed and averaged to obtain the period storage byte ratio; Compare the cycle storage byte ratios corresponding to all historical storage cycles, select the maximum cycle storage byte ratio and the minimum cycle storage byte ratio, and calculate the sum and average to obtain the characteristic byte ratio; The feature byte ratio is multiplied by the feature quantity ratio to obtain the storage ratio value; It can be understood that: the meaning of the storage ratio value is: reflecting the storage ratio of the cross-domain sound recognition model to the difference features in multiple historical storage cycles. Specifically, the storage ratio value is calculated by multiplying the feature byte ratio by the feature quantity ratio. Therefore, on the one hand, the feature byte ratio reflects the relative proportion of the difference feature byte size in the total bytes of all sound features. On the other hand, the feature quantity ratio reflects the relative proportion of the number of difference features in the number of all sound features. This is conducive to evaluating the storage ratio of the difference features by the cross-domain sound recognition model, so as to store the difference features of different sounds in a targeted manner, and provide data support for improving the accuracy of sound recognition by the cross-domain sound recognition model. Compare the storage share value with the storage share threshold as follows: If the storage ratio value is greater than the storage ratio threshold, it means that the storage occupancy of the difference feature is small, and a storage non-optimization signal is generated; If the storage ratio value is less than or equal to the storage ratio threshold, it means that the storage occupancy of the difference feature is large, and a storage optimization signal is generated; It should be noted that the storage ratio threshold is obtained as follows: The ratio of the number of all difference features in the feature difference degree sequence to the total number of time-frequency features corresponding to all sounds in different domains is calculated to obtain the feature difference quantity ratio; Obtain the byte of any difference feature in the feature difference degree sequence, calculate the ratio with the total bytes of all features corresponding to the sound, and perform summation and mean calculation to obtain the feature difference byte ratio; The storage ratio threshold is obtained by multiplying the feature difference quantity ratio and the feature difference byte ratio. Based on the storage of non-optimized signals, the difference feature values ​​corresponding to the difference features are compared in size, and are sorted and stored in descending order; The specific implementation method of the embodiment of the present invention is: based on the difference feature signal, the memory information of the difference features in multiple historical storage periods is obtained through PyTorch, and the storage ratio analysis is performed to obtain the storage ratio value. The storage ratio value reflects the storage ratio of the difference features of the cross-domain sound recognition model in multiple historical storage periods, which is conducive to evaluating the storage content ratio of the difference features of the cross-domain sound recognition model, so as to store the difference features of different sounds in a targeted manner, and provide data support for improving the accuracy of sound recognition by the cross-domain sound recognition model.

[0023] Example 3 like Figure 1 As shown, based on Example 1 and Example 2, the sound recognition model optimization method described in the embodiment of the present invention includes: Step 4: Based on the storage optimization signal, obtain the priority storage coefficient, and prioritize the storage of the difference features according to the priority storage coefficient to complete the storage optimization of the sound recognition model features; In some embodiments, the byte corresponding to the difference feature is extracted, and the ratio is calculated with the total amount of bytes corresponding to all the difference features to obtain the feature byte value, and all the difference features are sorted from large to small according to the feature byte values ​​corresponding to the difference features to obtain the feature byte storage sequence; Obtain the byte of any difference feature in the feature difference degree sequence, calculate the ratio with the total bytes of all features corresponding to the sound, and perform summation and mean calculation to obtain the feature difference byte ratio; Overlap and pair the feature byte storage sequence with the feature difference degree sequence, extract the difference features of the overlapped pairing, and obtain the corresponding feature difference value and feature difference byte ratio, calculate the ratio of the feature difference value and the feature difference byte ratio, and obtain the priority storage coefficient; Based on the priority storage coefficient, the priority storage coefficients corresponding to the difference features in the feature difference degree sequence are compared, and the difference features are preferentially stored in order from large to small; Dynamically count the feature difference byte ratios corresponding to the priority stored difference features to obtain the priority storage memory, until the priority storage memory is equal to the storage ratio threshold, and stop the difference feature storage work; The specific implementation method of the embodiment of the present invention is: based on the storage optimization signal, a priority storage coefficient is obtained, and the difference features are preferentially stored according to the priority storage coefficient. At the same time, the priority storage memory is dynamically obtained. When the priority storage memory is equal to the storage ratio threshold, the difference feature storage work is stopped, thereby solving the problem that when a cross-domain sound recognition model stores multiple difference features, the multiple difference features occupy a large memory and cannot be completely stored. Then, the storage is reasonably optimized to reduce the storage burden of the cross-domain sound recognition model, and the recognition accuracy of the cross-domain sound recognition model for sounds in different domains is improved.

[0024] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A cross-domain sound recognition model optimization method, characterized in that: include: Step 1: In the recognition and capture cycle, the features of sounds in different domains are imported into the cross-domain sound recognition model, and the time-frequency features are extracted to obtain multiple groups of time-frequency feature sequences; Step 2: Perform feature difference analysis on the time-frequency feature sequences corresponding to sounds in different domains to obtain zero rate difference values ​​and formant difference values, perform quantization processing to obtain difference features, and summarize to obtain a feature difference degree sequence; Step 3: In multiple historical storage cycles, obtain the feature byte ratio and feature quantity ratio, perform quantification processing, and compare them with the memory corresponding to all difference features in the feature difference degree sequence to generate a storage optimization signal; Step 4: Based on the storage optimization signal, obtain the priority storage coefficient, and complete the priority storage of the difference features according to the priority storage coefficient.

2. The cross-domain sound recognition model optimization method according to claim 1, characterized in that: The extraction process of time-frequency features is as follows: Divide the recognition and capture cycle into a number of capture time periods at equal time intervals; The time-frequency features are processed by short-time Wavelet transform to obtain the time-frequency feature values; The time-frequency feature values ​​corresponding to all capture time periods in the recognition capture period of different sounds are sorted according to the time series corresponding to the capture time periods to obtain multiple groups of time-frequency feature sequences.

3. The cross-domain sound recognition model optimization method according to claim 2, characterized in that: The process of obtaining the zero-rate difference value is as follows: A time-frequency feature sequence is randomly extracted as a feature comparison sequence, and the remaining time-frequency feature sequences after the extraction are combined with the feature comparison sequence one by one to form a comparison analysis group, thereby obtaining multiple comparison analysis groups; Randomly select a comparison analysis group, substitute the time-frequency feature values ​​corresponding to all time-frequency features into the zero-crossing rate formula, and obtain the zero-crossing rate corresponding to the time-frequency feature of each capture node; By obtaining the zero-crossing rates corresponding to all time-frequency features in the feature comparison sequence, the zero-crossing rates corresponding to all time-frequency features in the remaining time-frequency feature sequence after extraction are obtained, and the zero-crossing rate difference value is obtained through the formula.

4. The cross-domain sound recognition model optimization method according to claim 2, characterized in that: The resonance peak difference value is obtained as follows: Construct a feature comparison change curve, extract the time-frequency feature values ​​corresponding to all peaks, and integrate them according to the time series to obtain a comparison peak set; Construct a time-frequency feature change curve, extract the time-frequency feature values ​​corresponding to all peaks, and integrate them according to the time series to obtain a set of time-frequency feature peaks; Substitute the comparison peak set and the time-frequency feature peak set into the Euclidean distance formula to obtain the resonance peak difference value ; Extract the time-frequency feature values ​​corresponding to all troughs and integrate them according to the time series to obtain the comparison trough set; Extract the time-frequency feature values ​​corresponding to all troughs and integrate them according to the time series to obtain the time-frequency feature trough set; Substitute the comparison trough set and the time-frequency feature trough set into the Euclidean distance formula to obtain the resonance trough difference value ; The resonance peak difference Difference from resonance trough Perform sum calculation to obtain the resonance peak difference value.

5. The cross-domain sound recognition model optimization method according to claim 1, characterized in that: The feature difference degree sequence is obtained as follows: The zero-crossing rate difference value and the resonance peak difference value are summed up to obtain a characteristic difference value; If the feature difference value is greater than the feature difference threshold, a difference feature signal is generated, and the time-frequency feature of the generated difference feature signal is marked as a difference feature; Sort all difference features from large to small according to the corresponding feature difference values. Get the feature difference degree sequence.

6. The cross-domain sound recognition model optimization method according to claim 5, characterized in that: The feature quantity ratio is obtained as follows: Arbitrarily extract a historical storage period; Divide the historical storage cycle into several historical storage periods, obtain the number of stored difference features in the historical storage period, and calculate the ratio with the number of all features corresponding to the sound to obtain the unit storage quantity ratio; The unit storage quantity ratios corresponding to all historical storage periods in the historical storage cycle are summed and averaged to obtain the period storage quantity ratio; The period storage quantity ratios corresponding to all historical storage periods are compared, the maximum period storage quantity ratio and the minimum period storage quantity ratio are selected, and the sum and mean are calculated to obtain the characteristic quantity ratio.

7. The cross-domain sound recognition model optimization method according to claim 6, characterized in that: The characteristic byte ratio is obtained as follows: Obtain the bytes corresponding to the storage difference features during the historical storage period, and calculate the ratio with the total bytes of all features corresponding to the sound to obtain the unit storage byte ratio; The unit storage byte ratios corresponding to all historical storage periods in the historical storage cycle are summed and averaged to obtain the period storage byte ratio; Compare the cycle storage byte ratios corresponding to all historical storage cycles, select the maximum cycle storage byte ratio and the minimum cycle storage byte ratio, and calculate the sum and average. Get the characteristic byte ratio.

8. The cross-domain sound recognition model optimization method according to claim 1, characterized in that: The generation process of storage optimization signal is as follows: The feature byte ratio is multiplied by the feature quantity ratio to obtain the storage ratio value; If the storage ratio value is less than or equal to the storage ratio threshold, a storage optimization signal is generated.

9. The cross-domain sound recognition model optimization method according to claim 7, characterized in that: The priority storage coefficient is obtained as follows: Extract the bytes corresponding to the difference features, calculate the ratio with the total amount of bytes corresponding to all the difference features, obtain the feature byte value, and sort all the difference features from large to small according to the feature byte values ​​corresponding to the difference features, and obtain the feature byte storage sequence; Obtain the byte of any difference feature in the feature difference degree sequence, calculate the ratio with the total bytes of all features corresponding to the sound, and perform summation and mean calculation to obtain the feature difference byte ratio; The feature byte storage sequence is overlapped and paired with the feature difference degree sequence, the difference features of the overlapped pairing are extracted, and the corresponding feature difference value and feature difference byte ratio are obtained, and the feature difference value and feature difference byte ratio are calculated to obtain the priority storage coefficient.

10. The cross-domain sound recognition model optimization method according to claim 9, characterized in that: The difference features are stored preferentially according to the priority storage coefficient. The process is as follows: Compare the priority storage coefficients corresponding to the difference features in the feature difference degree sequence, and store the difference features in a priority order from large to small; The feature difference byte ratios corresponding to the priority stored difference features are dynamically counted to obtain the priority storage memory, until the priority storage memory is equal to the storage ratio threshold, and the difference feature storage work is stopped.

Images