A speech input system and method based on friction nanogeneration

Abstract

The application belongs to the technical field of speech recognition, and specifically discloses a speech input system and method based on friction nanogenerator, wherein the method comprises the following steps: collecting a triboelectric signal sequence, extracting signal intensity features and fluctuation mode analysis features from the sequence, and obtaining a preliminary signal fluctuation description; according to the preliminary signal fluctuation description, processing the part containing interference noise in the sequence by using a wavelet transform algorithm, and determining a pure signal sequence after removing the noise; if the signal intensity features in the pure signal sequence exceed a preset threshold, judging that it is a potential effective signal segment, and obtaining a time domain feature related to a pronunciation action sense and an action sense intensity index from the potential effective signal segment; through matching of the action sense intensity index and the fluctuation mode analysis, classifying the signals under the dynamic environment adaptation condition, and obtaining an effective signal candidate set after classification; and the application aims to solve the problem that the signal collection in the prior art is easily disturbed by noise, humidity and contact mode diversity under a dynamic environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of speech recognition, in particular to a speech input system and method based on friction nanogenerator. BACKGROUND

[0002] In the field of modern technology, speech input technology, as an important way of human-computer interaction, is widely used in intelligent devices, medical assistance and barrier-free communication scenarios, and its development has an undeniable value in improving the convenience of life and the inclusiveness of technology. Especially in some special environments, such as scenes where traditional microphones cannot be used, it is particularly urgent to explore new ways of speech input. Friction nanogenerator technology, as a new energy conversion and signal acquisition method, provides a new possibility for speech input. The tiny friction generated by body contact is converted into an electrical signal, which can capture the subtle movements of the user's pronunciation, opening up a new path for interaction in silent or low-voice environments. However, existing speech input methods still face many challenges in signal acquisition and processing, especially in complex environments. Many technologies are difficult to effectively distinguish between target signals and external interference, resulting in a decrease in input accuracy, especially when the user's pronunciation movements are weak or the environmental noise is strong, the system often cannot accurately identify effective speech features. This limitation makes it difficult for new technologies to achieve ideal results in practical applications, especially in scenarios that rely on body contact to generate signals, fluctuations in signal quality become an important bottleneck affecting system stability.

[0003] Further focusing on technical difficulties, the core challenge of friction nanogenerator in speech input lies in the evaluation and screening of signal quality. When the user pronounces, the contact between the oral cavity or tongue and the sensor will generate unstable frictional electric signals. The strength and fluctuation pattern of this signal is affected by the contact method, force, and even external humidity, making it difficult for the system to determine which signals are truly caused by speech actions and which are invalid interference or noise in a stationary state. The deeper problem is that this uncertainty in signal fluctuations directly affects the accuracy of subsequent signal analysis, for example, in some cases, the user's slight pronunciation may be misjudged as invalid input, while meaningless friction in the environment is treated as valid signal processing.

[0004] Therefore, how to accurately evaluate the fluctuation characteristics of frictional electric signals in a dynamically changing contact environment and screen out effective signal segments that truly reflect speech characteristics has become a key problem in improving the practicality and reliability of this technology. SUMMARY

[0005] The present application provides a speech input system and method based on friction nanogenerator, aiming to solve the problem of poor speech recognition accuracy caused by signal collection being easily disturbed by noise, humidity and contact method diversity in dynamic environment in the prior art.

[0006] To solve the above technical problems, the technical solution adopted by the present application is:

[0007] A voice input method based on friction nanogenerator, the method comprising: collecting a triboelectric signal sequence, extracting signal intensity features and fluctuation pattern analysis features therefrom to obtain a preliminary signal fluctuation description; according to the preliminary signal fluctuation description, using a wavelet transform algorithm to process the part containing interference noise in the sequence to determine the pure signal sequence after removing noise; if the signal intensity feature in the pure signal sequence exceeds a preset threshold, it is judged as a potential effective signal segment, and the time domain feature related to the pronunciation action sense is obtained from it to obtain the action sense intensity index; through the matching of the action sense intensity index and the fluctuation pattern analysis, the signals under the condition of dynamic environment adaptation are classified to obtain the classified effective signal candidate set; according to the classified effective signal candidate set, using a support vector machine algorithm to verify the voice feature capture, judging which paragraphs in the candidate set truly reflect the voice action, and determining the final effective signal segment set; if there are abnormal fluctuations related to the influence of external humidity in the final effective signal segment set, adjust the fluctuation pattern analysis of these paragraphs through an adaptive filtering algorithm to obtain the corrected signal segment set; according to the corrected signal segment set, cluster analysis is performed for the diversity of contact modes to obtain the clustering grouping result; through the comparison of the clustering grouping result and the signal evaluation standard, using the k-means algorithm to refine the boundary of the signals within the group to determine the optimized effective signal segment boundary description.

[0008] In an aspect of the present disclosure, the preliminary signal fluctuation description is obtained by collecting a triboelectric signal sequence, extracting signal intensity features and fluctuation pattern analysis features therefrom, comprising:

[0009] According to the above business content and the extracted related attributes, the following business solutions are generated, around the fluctuation description target of the triboelectric signal, combined with the interrelated attributes of signal collection, signal intensity, fluctuation pattern, feature extraction, data processing, sequence analysis, intensity analysis and pattern recognition, the following technical process steps are generated: real-time collection of triboelectric signals by sensor equipment to obtain original signal sequence data, stored as an initial data set;

[0010] According to the initial data set, using a segmentation processing method to divide the signal sequence into time windows to obtain a segmented signal segment set;

[0011] For the segmented signal segment set, calculate the signal intensity value in each segment, use the mean value calculation method to determine the intensity feature of each segment;

[0012] The fluctuation mode description is obtained by performing fluctuation mode analysis on the signal segment set, and using a fast Fourier transform method to extract fluctuation characteristics in a frequency domain;

[0013] If the frequency component in the fluctuation mode description exceeds a preset threshold range, the corresponding signal segment is subjected to secondary filtering processing, and an adjusted fluctuation characteristic set is obtained;

[0014] According to the adjusted fluctuation characteristic set and the intensity characteristic, a comprehensive signal fluctuation description model is constructed, and the fluctuation category of the signal is determined;

[0015] The final fluctuation description data is obtained by classifying and storing the output result of the comprehensive signal fluctuation description model.

[0016] In an aspect of the present disclosure, the pure signal sequence after removing noise is determined according to the preliminary signal fluctuation description by using a wavelet transform algorithm to process the part containing interference noise in the sequence, including:

[0017] The original data containing interference noise is obtained by initially collecting the signal sequence, and a signal data set to be processed is obtained;

[0018] The interference noise is separated by using a wavelet transform method for the signal data set to be processed, and the separated noise component and the preliminary pure signal are obtained;

[0019] According to the preliminary pure signal, a signal reconstruction operation is performed to recombine the separated signal data, and a reconstructed pure signal set is obtained;

[0020] The fluctuation characteristic extraction is implemented for the reconstructed pure signal set, the fluctuation change mode of the signal in the time domain is obtained, and the fluctuation characteristic description is determined;

[0021] If some segments in the fluctuation characteristic description exceed a preset threshold range, the corresponding segments are subjected to secondary smoothing processing, and the adjusted fluctuation characteristic data is obtained;

[0022] According to the adjusted fluctuation characteristic data, a signal fluctuation classification model is constructed, the category division of the signal fluctuation is obtained, and the final fluctuation attribution is determined;

[0023] The final fluctuation attribution result is stored in a structured manner, the database management tool is used to save the classification data, and the long-term queryable signal fluctuation archive is determined.

[0024] In an aspect of the present disclosure, if the signal intensity characteristic in the pure signal sequence exceeds a preset threshold, it is determined as a potential effective signal segment, and the time domain feature related to the pronunciation action feeling is obtained, and the action feeling intensity index is obtained, including:

[0025] By scanning the signal strength of the pure signal piece by piece, if it is detected that the signal strength of a piece exceeds a preset threshold, the piece is marked as an effective signal piece to obtain a preliminary screened signal piece set;

[0026] For the preliminary screened signal piece set, a feature extraction method is used to obtain time domain feature data related to the pronunciation action, and a signal piece range with strong action feeling is determined;

[0027] According to the signal piece range with strong action feeling, the corresponding intensity index value is calculated to obtain the quantized intensity data of each piece, and the significance ranking thereof in the overall signal is judged;

[0028] By classifying the quantized intensity data after the significance ranking, a pre-established support vector machine model is used to divide the signal pieces into different action intensity categories to obtain classified signal groups;

[0029] For the classified signal groups, the time domain distribution pattern of the signal pieces in each category is obtained to determine the correlation degree data thereof with the pronunciation action;

[0030] According to the correlation degree data, each signal group is marked with a priority, and if the correlation degree data is higher than a preset standard, the signal piece is marked as a high-priority signal piece to obtain a final priority processing signal set;

[0031] By structurally storing the final priority processing signal set, a database tool is used to save the classification and priority information to determine long-term queryable signal archive data.

[0032] In an aspect of the present disclosure, the classification of signals under dynamic environment adaptation conditions by matching the action intensity index and the fluctuation pattern analysis obtains a classified effective signal candidate set, including:

[0033] By preliminarily collecting signal data under a dynamic environment, a pre-established support vector machine model is used to classify and process the signals to obtain a preliminary classified signal set;

[0034] According to the preliminary classified signal set, the action intensity of each classification group is analyzed one by one, and if the intensity value of a certain classification group exceeds a preset threshold, the classification group is marked as a high-intensity signal group to obtain a high-intensity signal group set;

[0035] For the high-intensity signal group set, the fluctuation pattern data in each group is obtained, and the fluctuation patterns are compared by a pattern matching method to determine the signal group range consistent with the preset pattern;

[0036] According to the signal group range consistent with the preset mode, the environmental adaptability is detected, if the adaptability of a certain group in the dynamic environment is lower than the preset standard, it is eliminated, and a signal group set with qualified adaptability is obtained;

[0037] For the signal group set with qualified adaptability, a candidate screening operation is implemented, the signal data in each group is sorted, the signal data with high ranking is obtained, and a final effective candidate signal set is determined;

[0038] Through the structured storage of the final effective candidate signal set, the database tool is used to save the classification information and intensity data, and the signal archive data which can be queried for a long time is obtained;

[0039] According to the signal archive data which can be queried for a long time, a regular updating mechanism is implemented, if the dynamic environment changes, the reclassification operation of the signal set is triggered, and the updated effective candidate signal set is obtained.

[0040] In an aspect of the present disclosure, the voice feature capture is verified by using a support vector machine algorithm according to the classified effective signal candidate set, it is judged which paragraphs in the candidate set truly reflect the voice action, and a final effective signal paragraph set is determined, including:

[0041] Through data arrangement of the classified effective signal candidate set, the original data record of each signal paragraph in the candidate set is obtained, and a preliminary signal paragraph list is determined;

[0042] According to the preliminary signal paragraph list, the voice features of each signal paragraph are compared and analyzed by using a pre-established support vector machine model, and a signal paragraph grouping after feature analysis is obtained;

[0043] For the signal paragraph grouping after feature analysis, a matching detection of action standard is implemented, if the feature data of a certain signal paragraph does not conform to the action standard, it is eliminated, and a signal paragraph set conforming to the standard is obtained;

[0044] From the signal paragraph set conforming to the standard, the judgment basis data of each signal paragraph is obtained, the basis data is checked one by one, and a screened signal paragraph range is determined;

[0045] For the screened signal paragraph range, a signal verification operation is implemented, if it is detected that a certain signal paragraph appears abnormal in the verification process, it is marked as to be processed, and a signal paragraph list passed the verification is obtained;

[0046] According to the signal paragraph list passed the verification, the classification result data of each signal paragraph is arranged, the data is saved by using a structured storage tool, and a final signal paragraph archive is determined;

[0047] Through periodic detection on the final signal paragraph archive, if it is found that the signal paragraph data in the archive does not match the current environment, a data updating process is triggered to obtain an updated signal paragraph set.

[0048] In an aspect of the present disclosure, if there are abnormal fluctuations related to the influence of external humidity in the final effective signal segment set, the fluctuation pattern analysis of these paragraphs is adjusted through an adaptive filtering algorithm to obtain a corrected signal segment set, including:

[0049] For the final effective signal segment set, the humidity-related signal data is preliminarily scanned through a preset environmental monitoring tool, and if the scanning result shows that there are abnormal fluctuations, these signal segments are marked to obtain a marked signal segment list;

[0050] According to the marked signal segment list, an adaptive filtering algorithm is used to analyze and correct the fluctuation pattern of the marked signal segments one by one to obtain a corrected signal segment grouping;

[0051] For the corrected signal segment grouping, the fluctuation pattern of each signal segment is matched and detected with a preset standard through a data comparison tool, and if the fluctuation pattern of a certain signal segment does not match the standard, it is classified as to be processed to obtain a signal segment set that passes the matching;

[0052] According to the signal segment set that passes the matching, the environmental correlation data of each signal segment is obtained, and these data are stored in a structured manner through a data arrangement tool to determine an arranged signal segment archive;

[0053] For the arranged signal segment archive, the signal segments in the archive are compared with the current external environmental data through a periodic scanning tool, and if it is detected that a certain signal segment does not match the environmental data, it is marked as needing to be updated to obtain an updated signal segment list;

[0054] According to the updated signal segment list, the environmental data of the marked signal segments is re-acquired and recorded through a data acquisition module to obtain an updated signal segment data set;

[0055] For the updated signal segment data set, the updated data is merged with the original signal segment archive through a data integration tool to determine a final signal segment comprehensive archive.

[0056] In an aspect of the present disclosure, according to the corrected signal segment set, cluster analysis is performed for the diversity of contact mode changes to obtain a cluster grouping result, including:

[0057] For the association of the signal segment set and the correction data, the correction data is preliminarily arranged through a data screening tool to obtain an arranged signal segment basic archive;

[0058] According to the sorted signal segment basic file, in combination with the association of the contact mode and the diversity state, a classification tool is used for mode recognition of the contact mode of the signal segment, to determine a signal segment list after mode recognition;

[0059] In combination with the association of the clustering method and the grouping result, a clustering analysis tool is used for grouping processing of the signal segment, to obtain a signal segment set after grouping, in view of the signal segment list after mode recognition;

[0060] According to the signal segment set after grouping, in combination with the association of the feature analysis and the mode difference, feature data of the contact mode of each group is acquired, to determine a signal segment directory after feature extraction;

[0061] In combination with the association of the signal classification and the result division, if the feature data of a certain group does not match a preset standard, the group is marked, to obtain a signal segment group after marking, in view of the signal segment directory after feature extraction;

[0062] According to the signal segment group after marking, in combination with the association of the data combination and the set processing, a data integration module is used for secondary arrangement of the marked group, to determine a final signal segment classification file;

[0063] In combination with the association of the mode difference and the result division, if the contact mode feature of a certain classification is detected to be inconsistent with an expectation, a record tool is used to update the classification label of the classification, to obtain an updated signal segment classification list, in view of the final signal segment classification file.

[0064] In an aspect of the present disclosure, the k-means algorithm is used to refine the boundary of the signal in the group through comparison of the clustering grouping result and the signal evaluation standard, to determine an optimized effective signal segment boundary description, including:

[0065] Through the association of the clustering grouping and the signal data, a data screening module is used for preliminary classification of the signal data, to obtain a signal data set after classification;

[0066] According to the signal data set after classification, in combination with the association of the grouping result and the evaluation standard, a comparison tool is used for standard matching of the signal data set, to determine a signal data list after matching;

[0067] In combination with the association of the signal boundary and the grouping boundary, a boundary division tool is used for boundary labeling of the signal data, to obtain a signal data set after labeling, in view of the signal data list after matching;

[0068] According to the signal data set after labeling, in combination with the association of the optimization processing and the boundary description, a data arrangement module is used for descriptive induction of the signal boundary, to obtain an induced boundary description record;

[0069] For the inductive boundary description record, in combination with the association of data processing and evaluation criteria, if the boundary description of a certain segment of signal data is inconsistent with the preset criteria, the boundary is corrected through an adjustment tool to determine the corrected signal boundary file;

[0070] Through the corrected signal boundary file, in combination with the association of standard comparison and optimization processing, the signal boundary file is finally arranged by using an information integration module to obtain an arranged signal boundary data set;

[0071] For the arranged signal boundary data set, in combination with the association of signal evaluation and data processing, the signal boundary data set is stored by classification through a recording tool to obtain a stored signal boundary directory.

[0072] In another aspect of the present disclosure, the present disclosure also relates to a voice input system based on friction nanogeneration, which comprises:

[0073] A signal acquisition and feature extraction module is used to acquire a triboelectric signal sequence and extract signal intensity features and fluctuation pattern analysis features therefrom to obtain a preliminary signal fluctuation description;

[0074] A wavelet denoising module is used to process the part containing interference noise in the sequence according to the preliminary signal fluctuation description by using a wavelet transform algorithm to determine a pure signal sequence after removing noise;

[0075] A time domain feature extraction module is used to determine whether the signal intensity feature in the pure signal sequence exceeds a preset threshold value, and if it exceeds the threshold value, it is determined to be a potential effective signal segment, and the time domain feature related to the pronunciation action is obtained from it to obtain an action intensity index;

[0076] A dynamic classification module is used to classify the signals under dynamic environmental adaptation conditions by matching the action intensity index with the fluctuation pattern analysis to obtain a classified effective signal candidate set;

[0077] A support vector machine verification module is used to verify the voice feature capture by using a support vector machine algorithm according to the classified effective signal candidate set to determine the final effective signal segment set that truly reflects the voice action in the candidate set;

[0078] An adaptive filter correction module is used to determine whether there are abnormal fluctuations related to the influence of external humidity in the final effective signal segment set, and if there are abnormalities, the fluctuation pattern analysis of these segments is adjusted by using an adaptive filter algorithm to obtain a corrected signal segment set;

[0079] A clustering analysis module is used to perform clustering analysis on the diversity of contact methods according to the corrected signal segment set to obtain a clustering grouping result;

[0080] The clustering optimization module is used for refining the boundary of the signal in the grouping by comparing the grouping result with the signal evaluation standard by using the k-means algorithm, and determining the boundary description of the effective signal segment after optimization.

[0081] Compared with the prior art, the present application has the following beneficial effects:

[0082] The present application realizes effective recognition and signal optimization of voice action through multi-level signal processing and feature analysis. The present application firstly extracts preliminary features through signal intensity and fluctuation mode analysis, removes noise interference by using wavelet transform, and then screens potential effective signal segments and extracts time domain features, verifies the authenticity of voice action by combining support vector machine, and determines the effective signal segment set. For abnormal fluctuations caused by humidity, the present application uses adaptive filtering for correction, and optimizes the signal boundary through clustering analysis and k-means algorithm, and finally obtains accurate signal segment description. The technical effect of the present application lies in significantly improving the recognition accuracy and stability of voice action signals in complex environments, and providing a reliable solution for signal processing in dynamic environments. BRIEF DESCRIPTION OF DRAWINGS

[0083] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation to the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0084] Fig. 1 One of the flowcharts of a voice input method based on friction nanogenerator according to the present application.

[0085] Fig. 2 The second flowchart of a voice input method based on friction nanogenerator according to the present application.

[0086] Fig. 3 The third flowchart of a voice input method based on friction nanogenerator according to the present application. DETAILED DESCRIPTION

[0087] The present application will be further described below in conjunction with the embodiments. The described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0088] Please refer to Figs. 1-3 The present embodiment discloses a voice input method based on friction nanogenerator, which can specifically include:

[0089] Step S101, by collecting the triboelectric signal sequence, extracting the signal intensity feature and the fluctuation mode analysis feature, obtaining the preliminary signal fluctuation description.

[0090] According to the above business content and the extracted related attributes, the following business solutions are generated, around the fluctuation description target of the triboelectric signal, combined with the interrelated attributes of signal collection, signal intensity, fluctuation mode, feature extraction, data processing, sequence analysis, intensity analysis, pattern recognition, etc. The following technical process steps are generated: through the sensor device, the triboelectric signal is collected in real time, the original signal sequence data is obtained, and the initial data set is stored. According to the initial data set, the signal sequence is divided into time windows by using the segmentation processing method, and the segmented signal segment set is obtained. For the segmented signal segment set, the signal intensity value in each segment is calculated, the mean value calculation method is used to determine the intensity feature of each segment. Through the fluctuation mode analysis of the signal segment set, the fluctuation characteristics in the frequency domain are extracted by using the fast Fourier transform method, and the fluctuation mode description is obtained. If the frequency component in the fluctuation mode description exceeds the preset threshold range, the corresponding signal segment is filtered again, and the adjusted fluctuation feature set is obtained. According to the adjusted fluctuation feature set and the intensity feature, a comprehensive signal fluctuation description model is constructed to judge the fluctuation category of the signal. Through the classification storage of the output results of the comprehensive signal fluctuation description model, the final fluctuation description data is obtained.

[0091] Specifically, in the process of obtaining a preliminary signal fluctuation description by collecting a triboelectric signal sequence and extracting signal intensity and fluctuation pattern features, first, the sensor device can collect triboelectric signal data at a sampling rate of 1000 times per second. Assuming the collection duration is 10 seconds, 10000 data points are obtained, forming a time series data set. Subsequently, the data set is subjected to intensity feature extraction using signal processing algorithms, such as calculating the root mean square value of the signal. The specific method is to sum the square values of each data point, divide by the total number of data points, and then take the square root. Assuming the calculation result is an RMS value of 2.5 volts, it indicates that the overall signal intensity is at a medium level. Next, for fluctuation pattern feature analysis, the fast Fourier transform algorithm can be used to convert the time domain signal to the frequency domain signal, extract the main frequency components, and assume that the analysis result shows that the main frequency is 50 Hz and the secondary frequency is 100 Hz, indicating that the signal fluctuation has periodic characteristics. At the same time, the frequency spectrum energy distribution is calculated, and it is found that the energy proportion at 50 Hz is 60%, reflecting that this frequency is the dominant fluctuation pattern. To further describe the signal fluctuation, the standard deviation of the signal can be calculated to quantify the fluctuation amplitude. Assuming the standard deviation is 0.8 volts, combined with the RMS value and frequency spectrum analysis result, the preliminary signal fluctuation description is "medium intensity, obvious periodic fluctuation, and relatively stable fluctuation amplitude". To form a rigorous logical relationship, the above features can be associated with the device operating state. Assuming that the signal comes from a mechanical friction component, by comparing the standard deviation with historical data, assuming the historical standard deviation mean value is 0.5 volts, it is inferred that the current fluctuation amplitude is high, which may indicate that the friction component has a slight wear risk, thereby combining the signal analysis result with the business scenario to form a complete thought chain from data collection to feature extraction to application inference.

[0092] Step S102, according to the preliminary signal fluctuation description, the wavelet transform algorithm is used to process the part of the sequence containing interference noise to determine the pure signal sequence after removing noise.

[0093] The original data containing interference noise is obtained by initial acquisition of the signal sequence, and the signal data set to be processed is obtained. For the signal data set to be processed, the wavelet transform method is used to separate and process the interference noise, and the separated noise component and the preliminary pure signal are obtained. According to the preliminary pure signal, the signal reconstruction operation is performed, the separated signal data is recombined, and the reconstructed pure signal set is obtained. For the reconstructed pure signal set, the fluctuation characteristic extraction is implemented, the fluctuation change mode of the signal in the time domain is obtained, and the fluctuation characteristic description is determined. If some segments in the fluctuation characteristic description exceed the preset threshold range, the corresponding segments are subjected to secondary smoothing processing, and the adjusted fluctuation characteristic data is obtained. According to the adjusted fluctuation characteristic data, the signal fluctuation classification model is constructed, the category division of the signal fluctuation is obtained, and the final fluctuation attribution is determined. The final fluctuation attribution result is stored in a structured manner, the database management tool is used to save the classification data, and the long-term queryable signal fluctuation archive is determined.

[0094] Specifically, on the basis of the preliminary signal fluctuation description, the wavelet transform algorithm is used to process the interference noise that may exist in the triboelectric signal sequence to obtain a pure signal sequence, and the specific implementation method is as follows. First, it is assumed that there is a triboelectric signal time sequence data, the sampling rate is 500 times per second, the collection time length is 20 seconds, and there are 10000 data points in total, which contain background noise and random interference. Next, the Daubechies wavelet base (db8) in the wavelet transform algorithm is selected as the decomposition tool to decompose the signal into 5 layers of wavelet coefficients. By analyzing the energy distribution of each layer of coefficients, it is determined that the high-frequency part (the first layer and the second layer) mainly contains noise components, and the energy proportion is about 15% of the total energy. Then, the threshold processing method is set, and the soft threshold function is used to filter the high-frequency coefficients. It is assumed that the threshold is set to 0.3, the coefficients below the value are set to zero, and the coefficients above the value are reduced in proportion. After processing, the energy distribution is recalculated, and it is found that the high-frequency noise energy proportion is reduced to 3%. Finally, the signal is reconstructed by using the inverse wavelet transform to obtain the pure signal sequence. By comparing the signal-to-noise ratios of the signals before and after processing, it is assumed that the original signal SNR is 12.5 dB, and the processed signal SNR is improved to 18.7 dB, indicating that the noise interference is significantly reduced. In order to form a rigorous logical relationship, the pure signal sequence is combined with the device health monitoring business. It is assumed that the signal comes from a bearing friction component. By comparing the waveform with the historical pure signal, it is assumed that the historical SNR average is 20.0 dB. It is found that the current signal-to-noise ratio is slightly lower than the normal range, which may indicate that there is an initial abnormal wear hidden danger on the bearing surface, thereby providing data support for subsequent maintenance decision-making.

[0095] In step S103, if the signal intensity feature in the pure signal sequence exceeds the preset threshold, it is judged as a potential effective signal segment, and the time domain feature related to the pronunciation action feeling is obtained to obtain the action feeling intensity index.

[0096] By scanning the signal intensity of the pure signal piece by piece, if it is detected that the signal intensity of a certain piece exceeds the preset threshold, it is marked as an effective signal paragraph, and a preliminarily screened signal fragment set is obtained. For the preliminarily screened signal fragment set, a feature extraction method is used to obtain time domain feature data related to the pronunciation action, and the signal fragment range with strong action feeling is determined. According to the signal fragment range with strong action feeling, the corresponding intensity index value is calculated, the quantized intensity data of each fragment is obtained, and its significance ranking in the overall signal is judged. By classifying the quantized intensity data after significance ranking, using a pre-established support vector machine model, the signal fragments are divided into categories of different action intensity, and the classified signal groups are obtained. For the classified signal groups, the time domain distribution pattern of the signal paragraph in each category is obtained, and the correlation degree data of the signal paragraph with the pronunciation action is determined. According to the correlation degree data, each signal group is labeled with priority, and if the correlation degree data is higher than the preset standard, it is marked as a high-priority signal segment, and a final priority processing signal set is obtained. By structuring the final priority processing signal set, using a database tool to save the classification and priority information, long-term queryable signal archive data is determined.

[0097] Specifically, assuming that there is a processed pure signal sequence, the data comes from a vibration sensor related to human pronunciation action, the sampling rate is 1000 times per second, the collection time is 10 seconds, and there are a total of 10000 data points. First, the system automatically detects the signal intensity, the preset threshold is 0.5 volts, and by traversing the entire sequence, it identifies the continuous time period where the signal intensity exceeds the threshold. Assuming that 3 time periods meet the conditions, the total duration is 2.8 seconds, and it is determined that these are potential effective signal segments. Next, for these effective signal segments, the system uses a time domain feature extraction algorithm to calculate the root mean square value of the signal as the action intensity index. Assuming that the calculated RMS value is 0.72 volts, and the peak value feature is extracted, the peak value reaches 1.2 volts, and further analysis of the signal fluctuation frequency in the time period shows that it is mainly concentrated in the 5-10 Hz range, which meets the typical vibration characteristics of pronunciation action. Subsequently, the system compares the extracted action intensity index with the historical database, assuming that the RMS average of normal pronunciation action in the historical data is 0.68 volts, and the current value is slightly higher than the average level. Combined with the fluctuation frequency analysis, it is inferred that there may be potential characteristics of excessive force in pronunciation. In order to form a rigorous logical relationship, the analysis result is combined with the speech rehabilitation training business, and the system automatically generates an evaluation report, indicating that the current pronunciation action may have an abnormal force tendency, which can be used as a reference for subsequent training adjustment, thereby supporting the optimization design of individualized rehabilitation programs.

[0098] In step S104, the signals under the dynamic environment are classified by matching the action intensity index with the fluctuation pattern analysis, and a set of classified effective signal candidates is obtained.

[0099] Through preliminary collection of signal data under the dynamic environment, the signals are classified by using the pre-established support vector machine model, and a set of signals classified initially is obtained. According to the set of signals classified initially, the action intensity is analyzed one by one for each classification group. If the intensity value of a certain classification group exceeds the preset threshold, it is marked as a high-intensity signal group, and a set of high-intensity signal groups is obtained. For the set of high-intensity signal groups, the fluctuation pattern data in each group is obtained, the fluctuation patterns are compared by the pattern matching method, and the signal group range corresponding to the preset pattern is determined. According to the signal group range corresponding to the preset pattern, the environmental adaptability is detected. If the adaptability of a certain group under the dynamic environment is lower than the preset standard, it is removed, and a set of signal groups with qualified adaptability is obtained. For the set of signal groups with qualified adaptability, a candidate screening operation is implemented. The signal data in each group is sorted, the signal data at the top of the sorting is obtained, and a set of final effective candidate signals is determined. Through structured storage of the set of final effective candidate signals, the classification information and intensity data are saved by using the database tool, and signal archive data that can be queried for a long time is obtained. According to the signal archive data that can be queried for a long time, a regular update mechanism is implemented. If it is detected that the dynamic environment changes, a reclassification operation of the signal set is triggered, and an updated set of effective candidate signals is obtained.

[0100] Specifically, under the condition of dynamic environment adaptation, the system first acquires a segment of signal data from the vibration sensor related to human sound production. Assuming that the sampling rate is 800 times per second and the collection duration is 8 seconds, there are a total of 6400 data points. The signal is disturbed by environmental noise and needs to be classified. The system preprocesses the signal by using an adaptive filtering algorithm, filters out noise interference below 0.3 volts, retains the main signal components, and then extracts the action intensity index. Using a short-time energy analysis method, the average energy value in each 0.5-second window is calculated to obtain a set of energy distribution data, with the peak energy assumed to be 0.85 volts. Next, the system analyzes the signal fluctuation pattern, decomposes the signal using a wavelet transform algorithm, extracts characteristic components with a frequency range of 3-8 Hz, and finds that the main fluctuation period is 0.2 seconds, which is related to the dynamic characteristics of the sound production action. Based on this, the system classifies the signal according to the energy value and fluctuation period, sets the energy threshold to 0.6 volts and the cycle range to 0.15-0.25 seconds, classifies the signal segments that meet the conditions into the effective signal candidate set, and assumes that 4 candidate segments are finally selected, with a total duration of 1.9 seconds. To form a rigorous logical relationship, the system combines the classification results with the sound production behavior analysis service to automatically generate a feature matching report, which is used for subsequent sound production pattern evaluation in a dynamic environment, supports parameter adjustment of the adaptive training system, and ensures the application value of the classification results.

[0101] Step S105, according to the classified effective signal candidate set, the support vector machine algorithm is used to verify the speech feature capture, judge which paragraphs in the candidate set truly reflect the speech action, and determine the final effective signal segment set.

[0102] The original data records of each signal paragraph in the candidate set are obtained by data arrangement of the classified effective signal candidate set, and a preliminary signal paragraph list is determined. According to the preliminary signal paragraph list, the voice features of each signal paragraph are compared and analyzed by using a pre-established support vector machine model, and a signal paragraph grouping after feature analysis is obtained. For the signal paragraph grouping after feature analysis, the matching detection of action standards is implemented, if the feature data of a certain signal paragraph does not match the action standard, it is excluded, and a signal paragraph set meeting the standard is obtained. From the signal paragraph set meeting the standard, the judgment basis data of each signal paragraph is obtained, and the basis data is checked one by one to determine the range of the screened signal paragraph. For the screened signal paragraph range, signal verification operation is implemented, if it is detected that a certain signal paragraph appears abnormally in the verification process, it is marked as to be processed, and a signal paragraph list that passes the verification is obtained. According to the signal paragraph list that passes the verification, the classification result data of each signal paragraph is arranged, the data is saved by using a structured storage tool, and the final signal paragraph file is determined. Through regular detection of the final signal paragraph file, if it is found that the signal paragraph data in the file does not match the current environment, the data update process is triggered, and an updated signal paragraph set is obtained.

[0103] Specifically, in the processing of voice signals in a dynamic environment, the system further verifies and screens the classified effective signal candidate set to determine the final voice action related signal segment. First, the system extracts four signal segments from the candidate set, with a total duration of 2.2 seconds, assuming an average duration of 0.55 seconds for each signal segment, and a signal amplitude range of 0.4 to 0.9 volts. Then, the system uses a support vector machine algorithm to classify and verify the voice features of these signal segments. The specific method is to divide each signal segment into 0.1 second sub-segments, calculate the spectral features of each sub-segment, assume a frequency resolution of 0.5 Hz for spectral analysis, extract the energy distribution data of the main frequency components in the range of 2 to 10 Hz, and find that the energy concentration of two signal segments is more than 75%. Subsequently, the system classifies the feature vectors by using the support vector machine model, sets the classification boundary parameter to 0.7, selects the radial basis function as the kernel function, and calculates that the classification confidence of the two signal segments is 0.82 and 0.78 respectively, which exceeds the preset threshold of 0.75, while the confidence of the other two signal segments is only 0.62 and 0.58, which is judged as non-voice action related signal. Finally, the system classifies the two signal segments with confidence higher than the threshold into the final effective signal segment set, with a total duration of 1.1 seconds. To ensure logical rigor, the system associates the verification result with the voice behavior analysis business, automatically generates a classification verification log, records the feature values and confidence data of each signal segment, and provides them for subsequent voice pattern optimization module to support signal processing strategy adjustment in dynamic environment.

[0104] Step S106, if there are abnormal fluctuations related to the influence of external humidity in the final effective signal segment set, adjust the fluctuation pattern analysis of these segments through an adaptive filtering algorithm to obtain a corrected signal segment set.

[0105] For the final effective signal segment set, the signal data related to humidity is preliminarily scanned by a preset environmental monitoring tool. If the scanning result shows that there are abnormal fluctuations, the signal segments are marked to obtain a marked signal segment list. According to the marked signal segment list, the fluctuation pattern of the marked signal segments is analyzed and corrected one by one using an adaptive filtering algorithm to obtain a corrected signal segment grouping. For the corrected signal segment grouping, the fluctuation pattern of each signal segment is matched and detected with a preset standard by a data comparison tool. If the fluctuation pattern of a certain signal segment does not conform to the standard, it is classified as to be processed to obtain a signal segment set that passes the matching. According to the signal segment set that passes the matching, the environmental correlation data of each signal segment is obtained, and these data are stored in a structured manner by a data arrangement tool to determine an arranged signal segment archive. For the arranged signal segment archive, the signal segments in the archive are compared with the current external environmental data by a periodic scanning tool. If it is detected that a certain signal segment is inconsistent with the environmental data, it is marked as to be updated to obtain a to-be-updated signal segment list. According to the to-be-updated signal segment list, the environmental data of the marked signal segments is re-acquired and recorded by a data acquisition module to obtain an updated signal segment data set. For the updated signal segment data set, the updated data is merged with the original signal segment archive by a data integration tool to determine the final signal segment comprehensive archive.

[0106] Specifically, in the field of speech signal processing in dynamic environments, the system detects and corrects abnormal fluctuations related to external humidity that may exist in the final set of valid signal segments to ensure signal quality. First, the system scans the final set of valid signal segments for environmental parameters. Assuming the set contains 3 signal segments with a total duration of 1.5 seconds and an average duration of 0.5 seconds per signal segment, the system obtains environmental humidity data through the built-in humidity sensor interface. Assuming the current humidity value is 78%, and combining historical data analysis, it is found that when the humidity exceeds 75%, the signal fluctuation amplitude increases by about 15%. Next, the system detects fluctuations in each signal segment. The specific method is to divide the signal segment into 0.2-second subintervals, calculate the peak and trough values of each subinterval, and assume that the fluctuation amplitude of the first signal segment reaches 0.3 volts, exceeding the threshold of 0.1 volts in the normal range, and is determined as abnormal fluctuation. Subsequently, the system starts an adaptive filtering algorithm to adjust the abnormal signal segment. Using the least mean square error algorithm, the filter order is set to 8 and the iteration step is set to 0.01. By smoothing the time domain data of the signal segment, the fluctuation amplitude is reduced to within 0.12 volts, while preserving the main frequency components of the original signal in the 3 to 8 Hz range. Finally, the system generates a corrected signal segment set with a total duration of 1.5 seconds, and associates the fluctuation data before and after correction with the environmental humidity value, automatically storing it in the signal quality monitoring database for subsequent environmental adaptability optimization module calls to support signal processing strategy adjustments under different humidity conditions.

[0107] Step S107, according to the corrected signal segment set, cluster analysis is performed for the diversity of contact mode changes, and the clustering grouping result is obtained.

[0108] For the association of the signal segment set and the correction data, the correction data is preliminarily arranged through a data screening tool to obtain a signal segment basic file after arrangement. According to the signal segment basic file after arrangement, in combination with the association of the contact mode and the diversity state, a classification tool is used to recognize the mode of the signal segment, and a signal segment list after mode recognition is determined. For the signal segment list after mode recognition, in combination with the association of the clustering method and the grouping result, a clustering analysis tool is used to group the signal segment, and a signal segment set after grouping is obtained. According to the signal segment set after grouping, in combination with the association of the feature analysis and the mode difference, the contact mode feature data of each group is obtained, and a signal segment directory after feature extraction is determined. For the signal segment directory after feature extraction, in combination with the association of the signal classification and the result division, if the feature data of a certain group does not match the preset standard, the group is marked, and a signal segment group after marking is obtained. According to the signal segment group after marking, in combination with the association of the data combination and the set processing, a data integration module is used to secondarily arrange the marked group, and a final signal segment classification file is determined. For the final signal segment classification file, in combination with the association of the mode difference and the result division, if it is detected that the contact mode feature of a certain classification does not match the expectation, a record tool is used to update the classification label, and an updated signal segment classification list is obtained.

[0109] Specifically, in the field of speech signal processing in dynamic environments, the system performs clustering analysis on the corrected signal segment set based on the diversity of contact methods to obtain grouping results. First, the system extracts features from the corrected signal segment set. Assuming that the set contains 5 signal segments with a total duration of 2.5 seconds and an average duration of 0.5 seconds for each signal segment, by analyzing the time domain and frequency domain features of each signal segment, the contact method related parameters are extracted, such as the average energy value of the signal is 2.3 milliwatts, and the frequency distribution range is between 5 and 10 hertz. Subsequently, the system uses the K-means clustering algorithm for grouping, sets the number of cluster centers to 3, initializes the center points based on random sampling of signal energy distribution, and sets the upper limit of iteration times to 50 times. In each iteration, the Euclidean distance of each signal segment to the center point is calculated. Assuming that after the first iteration, the distance from signal segment 1 to center point 1 is 0.8, and the distance from signal segment 2 to center point 2 is 1.2, the system reassigns the signal segment attribution based on the distance and updates the center point position. After 30 iterations, convergence is achieved, and 3 clustering groups are obtained, including 2 signal segments in the first group with an average energy of 2.1 milliwatts, 2 signal segments in the second group with an average energy of 2.5 milliwatts, and 1 signal segment in the third group with an average energy of 2.0 milliwatts. Then, the system verifies the features of the clustering results by calculating the within-group variance and between-group variance. Assuming that the within-group variance is 0.15 and the between-group variance is 0.85, the significance of the clustering results is verified, and the grouping is confirmed to be reasonable. Finally, the system performs correlation analysis on the clustering grouping results and the physical characteristics of the contact method, combines the contact mode data stored in the historical database, generates classification labels, such as the first group corresponds to contact method A, the second group corresponds to contact method B, and the third group corresponds to contact method C, and automatically stores the results to the signal classification database, providing reference for subsequent signal processing strategy optimization.

[0110] In step S108, the boundaries of the signals within the grouping are refined by comparing the clustering grouping results with the signal evaluation criteria using the k-means algorithm to determine the optimized effective signal segment boundary description.

[0111] The signal data set after classification is obtained by adopting a data screening module to preliminarily classify the signal data through the association of the clustering grouping and the signal data. According to the classified signal data set, the matching signal data list is determined by using a comparison tool to match the signal data set in combination with the association of the grouping result and the evaluation standard. In combination with the association of the signal boundary and the grouping boundary, the signal data is marked with a boundary by using a boundary division tool for the matching signal data list, so as to obtain the signal data set after marking. According to the signal data set after marking, the descriptive induction of the signal boundary is performed by using a data arrangement module in combination with the association of the optimization processing and the boundary description, so as to obtain the boundary description record after induction. In combination with the association of the data processing and the evaluation standard, the boundary of the signal data is corrected by using an adjustment tool if the boundary description of the signal data is inconsistent with the preset standard for the boundary description record after induction, so as to determine the signal boundary file after correction. In combination with the association of the standard comparison and the optimization processing, the signal boundary file is finally arranged by using an information integration module for the signal boundary file after correction, so as to obtain the signal boundary data set after arrangement. In combination with the association of the signal evaluation and the data processing, the signal boundary data set is stored by using a recording tool for the signal boundary data set after arrangement, so as to obtain the signal boundary directory after storage.

[0112] Specifically, in the field of speech signal processing in dynamic environments, the system compares the clustering grouping results with signal evaluation criteria, refines the signal boundaries within the grouping using the K-means algorithm, and finally determines the optimized effective signal segment boundary description. The system first retrieves the existing clustering grouping results from the database, assuming that it contains four signal segments, belonging to two groupings, the first group has three signal segments, with an average duration of 0.6 seconds, and the second group has one signal segment, with a duration of 0.4 seconds. Then, these signal segments are compared with the preset signal evaluation criteria, which include a signal duration threshold of 0.5 seconds and an energy fluctuation range of 1.5 to 3.0 milliwatts. Through comparison, it is found that one signal segment in the first group has a duration of 0.45 seconds, which does not meet the threshold, and the energy fluctuation is 1.2 milliwatts, which needs further refinement. Next, the system starts the K-means algorithm to adjust the boundaries, sets the number of clustering centers to 2, initializes the center points based on the signal duration distribution, and sets the upper limit of iterations to 20 times. In the first iteration, the distance of each signal segment to the center point is calculated, assuming that the distance of signal segment 1 to center point 1 is 0.3 and the distance of signal segment 2 to center point 2 is 0.5. The system reassigns the signal segment attribution and updates the center points based on the distance. After 15 iterations, the algorithm converges, and the refined first group signal segment boundary is adjusted to contain two signal segments, with an average duration of 0.55 seconds and an energy fluctuation of 1.8 milliwatts, meeting the evaluation criteria. The system then verifies the adjusted boundary, calculates the consistency of the signal duration within the boundary, and assumes that the consistency index is 0.9, reaching the preset threshold of 0.85, confirming that the boundary optimization is reasonable. Finally, the system automatically updates the optimized signal segment boundary description to the signal processing database and links with the subsequent signal classification module to ensure the continuity of the signal segment in different processing stages, forming a complete logical chain from boundary refinement to classification application.

[0113] The application provides a speech input system based on friction nanogenerator, mainly comprising:

[0114] A signal acquisition and feature extraction module is used to extract signal intensity features and fluctuation pattern analysis features from the collected triboelectric signal sequence to obtain a preliminary signal fluctuation description.

[0115] A wavelet denoising module is used to process the part of the sequence containing interference noise using a wavelet transform algorithm based on the preliminary signal fluctuation description to determine the pure signal sequence after removing noise.

[0116] A time domain feature extraction module is used to determine whether the signal intensity feature in the pure signal sequence exceeds the preset threshold. If it exceeds the threshold, it is determined as a potential effective signal segment, and the time domain feature related to the pronunciation action sense is obtained to obtain the action sense intensity index.

[0117] The dynamic classification module is configured to classify the signals under the dynamic environment adaptation condition through matching of the action intensity index and the fluctuation pattern analysis, and obtain a classified effective signal candidate set;

[0118] The support vector machine verification module is configured to verify the speech feature capture by using a support vector machine algorithm according to the classified effective signal candidate set, judge the segments in the candidate set that truly reflect the speech action, and determine a final effective signal segment set;

[0119] The adaptive filter correction module is configured to judge whether there is an abnormal fluctuation related to the influence of external humidity in the final effective signal segment set, and if there is an abnormality, adjust the fluctuation pattern analysis of these segments by using an adaptive filter algorithm to obtain a corrected signal segment set.

[0120] The clustering analysis module is configured to perform clustering analysis on the diversity of the contact mode change according to the corrected signal segment set, and obtain a clustering grouping result.

[0121] The clustering optimization module is configured to refine the boundaries of the signals in the grouping by using a k-means algorithm through comparison of the clustering grouping result and a signal evaluation standard, and determine an optimized effective signal segment boundary description.

[0122] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A voice input method based on triboelectric nanogenerators, characterized in that, The method includes: By collecting triboelectric signal sequences, signal intensity features and fluctuation pattern analysis features are extracted from them to obtain a preliminary description of signal fluctuation. Based on the preliminary description of signal fluctuations, wavelet transform algorithm is used to process the interference noise in the sequence to determine the clean signal sequence after noise removal. If the signal intensity characteristics in a pure signal sequence exceed a preset threshold, it is judged as a potentially effective signal segment, and the time-domain characteristics related to the sense of vocalization are obtained from it to obtain the sense of vocalization intensity index. By matching the motion intensity index with the fluctuation pattern analysis, the signals under dynamic environmental adaptation conditions are classified to obtain a set of effective signal candidates after classification. Based on the classified effective signal candidate set, the support vector machine algorithm is used to verify the captured speech features, determine the segments in the candidate set that truly reflect speech actions, and determine the final set of effective signal segments. If there are abnormal fluctuations related to the influence of external humidity in the final set of valid signal segments, the fluctuation pattern analysis of the segments is adjusted by an adaptive filtering algorithm to obtain a corrected set of signal segments. Based on the corrected signal segment set, cluster analysis is performed to assess the diversity of contact mode variations, resulting in cluster grouping results. By comparing the clustering results with the signal evaluation criteria, the k-means algorithm is used to refine the boundaries of the signals within the groups and determine the optimized effective signal segment boundary description. If the signal intensity characteristics in the pure signal sequence exceed a preset threshold, it is determined to be a potentially valid signal segment, and the time-domain features related to the vocalization motion are obtained from it to obtain a motion intensity index, including: By scanning the signal strength of the clean signal segment by segment, if a signal strength exceeds a preset threshold, it is marked as a valid signal segment, thus obtaining a preliminary set of filtered signal segments. For the initially screened set of signal segments, feature extraction methods are used to obtain time-domain feature data related to vocalization actions, and to determine the range of signal segments with strong sense of movement. Based on the range of signal segments with strong motion, calculate the corresponding intensity index value, obtain the quantized intensity data of each segment, and determine its salience ranking in the overall signal; By classifying the quantized intensity data after saliency ranking, a pre-established support vector machine model is used to divide the signal segments into categories with different motion intensity, resulting in classified signal groups. For the classified signal groups, the time domain distribution pattern of signal segments in each category is obtained to determine their correlation with the vocalization action. Based on the correlation data, each signal group is assigned a priority label. If the correlation data is higher than the preset standard, it is marked as a high-priority signal segment, thus obtaining the final set of priority-processed signals. By structuring and storing the final set of priority signals, and using database tools to save classification and priority information, long-term queryable signal archive data is determined.

2. The voice input method based on triboelectric nanogenerator according to claim 1, characterized in that, The process involves acquiring triboelectric signal sequences, extracting signal intensity features and fluctuation pattern analysis features from them, and obtaining a preliminary description of signal fluctuation, including: The triboelectric signal is collected in real time by sensor devices to obtain the original signal sequence data and store it as an initial dataset; Based on the initial dataset, the signal sequence is divided into time windows using a segmentation method to obtain a set of segmented signal segments. For the segmented signal fragment set, the signal strength value within each fragment is calculated, and the intensity characteristics of each fragment are determined by the mean value calculation method. By performing wave pattern analysis on a set of signal segments, the wave characteristics in the frequency domain are extracted using the Fast Fourier Transform method to obtain a wave pattern description. If the frequency components in the fluctuation pattern description exceed the preset threshold range, the corresponding signal segment is subjected to secondary filtering to obtain the adjusted fluctuation feature set. Based on the adjusted set of fluctuation characteristics and intensity characteristics, a comprehensive signal volatility description model is constructed to determine the volatility category of the signal; By classifying and storing the output results of the comprehensive signal volatility description model, the final volatility description data can be obtained.

3. The voice input method based on triboelectric nanogenerator according to claim 1, characterized in that, The step of processing the interference noise portion of the sequence using a wavelet transform algorithm based on the preliminary signal fluctuation description to determine the clean signal sequence after noise removal includes: By initially acquiring the signal sequence, the raw data containing interference noise is obtained, resulting in the signal dataset to be processed. For the signal dataset to be processed, wavelet transform is used to separate the interference noise, and the separated noise components and the preliminary clean signal are obtained. Based on the initial clean signal, a signal reconstruction operation is performed to reassemble the separated signal data and obtain a reconstructed set of clean signals. For the reconstructed set of clean signals, wave characteristics are extracted to obtain the wave change patterns of the signals in the time domain and determine the wave characteristic description. If some segments in the fluctuation characteristic description exceed the preset threshold range, the corresponding segments are subjected to secondary smoothing to obtain the adjusted fluctuation characteristic data. Based on the adjusted volatility characteristic data, a signal volatility classification model is constructed to obtain the category classification of signal volatility and determine the final volatility attribution. By structuring and storing the final volatility attribution results, and using database management tools to save the categorized data, a long-term queryable signal volatility archive is established.

4. The voice input method based on triboelectric nanogenerator according to claim 1, characterized in that, The process involves matching motion intensity indices with wave pattern analysis to classify signals under dynamic environmental adaptation conditions, resulting in a set of effective signal candidates, including: By initially collecting signal data in a dynamic environment, a pre-established support vector machine model is used to classify the signals, resulting in a pre-classified signal set. Based on the initial classification of the signal set, the intensity of the action is analyzed one by one for each classification group. If the intensity value of a certain classification group exceeds the preset threshold, it is marked as a high-intensity signal group, and a set of high-intensity signal groups is obtained. For a set of high-intensity signal groups, acquire the fluctuation pattern data within each group, compare the fluctuation patterns using a pattern matching method, and determine the range of signal groups that match the preset patterns; Based on the range of signal groups that match the preset mode, environmental adaptability testing is performed. If the adaptability of a certain group in a dynamic environment is found to be lower than the preset standard, it is removed, and a set of signal groups with qualified adaptability is obtained. For the set of signal groups that meet the adaptability requirements, a candidate screening operation is performed. By sorting the signal data within each group, the top-ranked signal data is obtained, and the final set of valid candidate signals is determined. By structuring and storing the final set of effective candidate signals, and using database tools to save classification information and intensity data, signal archive data that can be queried for a long time is obtained. Based on long-term queryable signal archive data, a periodic update mechanism is implemented. If a change in the dynamic environment is detected, a reclassification operation of the signal set is triggered to obtain an updated set of valid candidate signals.

5. The voice input method based on triboelectric nanogenerator according to claim 1, characterized in that, The step involves using a support vector machine algorithm to verify the captured speech features based on the classified candidate set of valid signals, determining the segments in the candidate set that truly reflect speech actions, and finally identifying the final set of valid signal segments, including: By organizing the data of the classified effective signal candidate set, the original data records of each signal segment in the candidate set are obtained, and a preliminary list of signal segments is determined. Based on the preliminary list of signal segments, a pre-established support vector machine model is used to compare and analyze the speech features of each signal segment, resulting in signal segment grouping after feature analysis. For the signal segments grouped after feature analysis, a matching detection of action standards is performed. If the feature data of a certain signal segment does not match the action standard, it is removed, and a set of signal segments that meet the standard is obtained. From the set of signal segments that meet the standards, obtain the judgment criteria data for each signal segment, and determine the range of filtered signal segments by checking the criteria data one by one. For the filtered signal segment range, a signal verification operation is performed. If an anomaly is detected in a signal segment during the verification process, it is marked as pending processing, and a list of verified signal segments is obtained. Based on the list of verified signal segments, the classification results data for each signal segment are organized, and the data is saved using a structured storage tool to determine the final signal segment archive. By periodically checking the final signal segment archive, if the signal segment data in the archive is found to be inconsistent with the current environment, a data update process is triggered to obtain an updated set of signal segments.

6. The voice input method based on triboelectric nanogenerator according to claim 1, characterized in that, If the final set of valid signal segments contains abnormal fluctuations related to external humidity, then an adaptive filtering algorithm is used to adjust the fluctuation pattern analysis of the segments to obtain a corrected set of signal segments, including: For the final set of valid signal segments, a preliminary scan of humidity-related signal data is performed using a preset environmental monitoring tool. If the scan results show abnormal fluctuations, these signal segments are marked to obtain a list of marked signal segments. Based on the list of marked signal segments, an adaptive filtering algorithm is used to analyze and correct the fluctuation patterns of the marked signal segments one by one, resulting in corrected signal segment groups. For the corrected signal segment groups, the fluctuation pattern of each signal segment is matched with the preset standard using a data comparison tool. If the fluctuation pattern of a certain signal segment does not match the standard, it is classified as pending processing, and a set of matched signal segments is obtained. Based on the set of matched signal segments, obtain the environmental association data for each signal segment, and use data processing tools to store this data in a structured manner to determine the processed signal segment archive; For the organized signal segment files, the signal segments in the files are compared with the current external environmental data through a periodic scanning tool. If a signal segment is found to be inconsistent with the environmental data, it is marked as needing to be updated, and a list of signal segments to be updated is obtained. Based on the list of signal segments to be updated, the environmental data of the marked signal segments is reacquired and recorded through the data acquisition module to obtain the updated signal segment dataset. For the updated signal segment dataset, the updated data is merged with the original signal segment archives using a data integration tool to determine the final comprehensive signal segment archive.

7. The voice input method based on triboelectric nanogenerator according to claim 1, characterized in that, The step involves performing cluster analysis based on the corrected signal segment set to assess the diversity of contact mode variations, resulting in cluster grouping results, including: To establish the relationship between the signal segment set and the calibration data, the calibration data is initially organized using a data filtering tool to obtain the basic file of the organized signal segments. Based on the compiled basic data of signal segments, and combined with the relationship between contact methods and diversity patterns, a classification tool is used to perform pattern recognition on the contact methods of signal segments to determine the list of signal segments after pattern recognition. Based on the list of signal segments after pattern recognition, and combining the correlation between clustering methods and grouping results, the signal segments are grouped using clustering analysis tools to obtain a set of grouped signal segments. Based on the grouped signal segment set, and combined with the correlation between feature analysis and mode differences, the contact mode feature data of each group is obtained, and the signal segment directory after feature extraction is determined. For the signal segment catalog after feature extraction, and combining the correlation between signal classification and result division, if the feature data of a certain group does not match the preset standard, it is marked to obtain the marked signal segment group. Based on the marked signal segment groups, and combined with the relationship between data combination and set processing, the marked groups are reorganized a second time through the data integration module to determine the final signal segment classification file. For the final signal segment classification file, combining the correlation between the differences in the method and the result classification, if the contact method characteristics of a certain category are detected to be inconsistent with the expectations, its classification label is updated through the recording tool to obtain an updated signal segment classification list.

8. The voice input method based on triboelectric nanogenerator according to claim 1, characterized in that, The process involves comparing the clustering results with signal evaluation criteria, using the k-means algorithm to refine the boundaries of signals within each group, and determining the optimized effective signal segment boundary description, including: By associating clustering with signal data, a data filtering module is used to perform preliminary classification of the signal data, resulting in a classified signal dataset. Based on the classified signal dataset, and combining the grouping results with the evaluation criteria, the signal dataset is matched against the criteria using a comparison tool to determine the matched signal data list. For the matched list of signal data, and combining the association between signal boundaries and group boundaries, a boundary delineation tool is used to mark the boundaries of the signal data to obtain the marked set of signal data. Based on the labeled signal data set, and combined with the correlation between optimization processing and boundary description, the signal boundaries are descriptively summarized through the data processing module to obtain the summarized boundary description record. For the summarized boundary description records, combined with the correlation between data processing and evaluation standards, if the boundary description of a certain segment of signal data is inconsistent with the preset standard, the boundary is corrected by adjustment tools to determine the corrected signal boundary file. By combining the revised signal boundary profile with the correlation between standard comparison and optimization processing, the information integration module is used to finally organize the signal boundary profile, resulting in the organized signal boundary dataset. For the organized signal boundary dataset, and combining the relationship between signal evaluation and data processing, the signal boundary dataset is classified and stored using a recording tool to obtain the stored signal boundary directory.

9. A voice input system based on triboelectric nanogenerators, characterized in that, The system includes: The signal acquisition and feature extraction module is used to acquire triboelectric signal sequences, extract signal intensity features and fluctuation pattern analysis features from them, and obtain a preliminary description of signal fluctuation. The wavelet denoising module is used to process the interference noise in the sequence based on the initial signal fluctuation description and the wavelet transform algorithm to determine the clean signal sequence after noise removal. The time-domain feature extraction module is used to determine whether the signal intensity features in a clean signal sequence exceed a preset threshold. If they exceed the threshold, they are judged as potentially effective signal segments, and the time-domain features related to the sense of pronunciation action are obtained from them to obtain the sense of action intensity index. The dynamic classification module is used to classify signals under dynamic environmental adaptation conditions by matching motion intensity index with fluctuation pattern analysis, and obtain a set of effective signal candidates after classification. The support vector machine verification module is used to verify the captured speech features using the support vector machine algorithm based on the classified effective signal candidate set, determine the segments in the candidate set that truly reflect speech actions, and determine the final set of effective signal segments. The adaptive filtering correction module is used to determine whether there are abnormal fluctuations related to the influence of external humidity in the final effective signal segment set. If there are abnormalities, the fluctuation pattern analysis of these segments is adjusted through the adaptive filtering algorithm to obtain the corrected signal segment set. The clustering analysis module is used to perform clustering analysis on the diversity of contact mode changes based on the corrected signal segment set, and obtain clustering results; The clustering optimization module is used to refine the boundaries of signals within a group by comparing the clustering results with the signal evaluation criteria and using the k-means algorithm to determine the optimized effective signal segment boundary description. The time-domain feature extraction module includes: By scanning the signal strength of the clean signal segment by segment, if a signal strength exceeds a preset threshold, it is marked as a valid signal segment, thus obtaining a preliminary set of filtered signal segments. For the initially screened set of signal segments, feature extraction methods are used to obtain time-domain feature data related to vocalization actions, and to determine the range of signal segments with strong sense of movement. Based on the range of signal segments with strong motion, calculate the corresponding intensity index value, obtain the quantized intensity data of each segment, and determine its salience ranking in the overall signal; By classifying the quantized intensity data after saliency ranking, a pre-established support vector machine model is used to divide the signal segments into categories with different motion intensity, resulting in classified signal groups. For the classified signal groups, the time domain distribution pattern of signal segments in each category is obtained to determine their correlation with the vocalization action. Based on the correlation data, each signal group is assigned a priority label. If the correlation data is higher than the preset standard, it is marked as a high-priority signal segment, thus obtaining the final set of priority-processed signals. By structuring and storing the final set of priority signals, and using database tools to save classification and priority information, long-term queryable signal archive data is determined.

Images