Voice data classification storage method and system based on Internet of Things platform

By preprocessing and classifying voice data, selecting suitable storage channels and memory for storage, the problem of low storage efficiency and transmission blockage in the Internet of Things platform is solved, and efficient data management and storage is achieved.

CN118410201BActive Publication Date: 2024-11-12深圳市双银科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410356110.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-27
Publication Date
2024-11-12
Estimated Expiration
2044-03-27

AI Technical Summary

Technical Problem

The prior art fails to store finely based on the size and category attributes of voice data in the Internet of Things platform, resulting in low storage efficiency and retrieval efficiency, and traffic jamming is prone to occur during data transmission.

Method used

By preprocessing and class analysis of voice data, selecting suitable storage channels and memory for storage, ensuring the integration and classification of data segments, selecting storage channels with a channel remaining capacity greater than the storage volume, and storing the data in the corresponding category attributes according to the category attributes in the sub-memory.

Benefits of technology

It improves the efficiency of voice data management and query, avoids traffic jams in data transmission, and ensures storage stability and durability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118410201B_ABST
    Figure CN118410201B_ABST
Patent Text Reader

Abstract

The present invention discloses a voice data classification storage method and system based on an Internet of Things platform, relates to the field of voice data technology, and is intended to solve the problems of inaccurate voice data classification and poor storage security. The present invention confirms the storage capacity of voice data to be stored according to the number of data segments, and by classifying and integrating the data segments, it is possible to more efficiently manage and access data, improve the efficiency of data processing and query, select a storage channel whose remaining channel capacity is greater than the storage capacity of voice data to be stored as the first storage channel, further improve the efficiency of voice data to be stored when storing and transmitting, and use a memory whose remaining storage capacity is greater than the storage capacity of voice data to be stored as the final memory for voice data to be stored, further ensuring the stability and durability of voice data to be stored when storing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of voice data technology, and in particular to a voice data classification storage method and system based on an Internet of Things platform. Background Art

[0002] The voice data of the IoT platform mainly comes from various smart terminal devices, such as smart home, smart car, smart customer service, etc. These devices are usually equipped with speech recognition and speech synthesis technology, which can collect user voice information and transmit it to the IoT platform for processing and analysis.

[0003] The Chinese patent with the announcement number CN116975355B discloses a voice data classification storage method for an IoT voice platform, which mainly modulates the voice signal of the first voice data interaction unit through the voice signal modulation module in the IoT voice platform to obtain multiple modulated interactive voice data, and the multiple modulated interactive voice data have corresponding multiple voice signal restoration identifiers; the first voice data storage chain is obtained by chain storage of the multiple modulated interactive voice data and the multiple voice signal restoration identifiers, and the first voice data storage chain is stored in the distributed voice data storage module of the IoT voice platform. Although the above patent solves the problem of voice data storage, the following problems still exist in actual operation:

[0004] 1. The corresponding storage space is not selected according to the size of the voice data, and the voice data is not stored more finely according to its category attributes, which leads to reduced storage and subsequent retrieval efficiency.

[0005] 2. Effective data transmission is not performed according to the storage capacity of voice data, resulting in traffic congestion during data transmission to the storage.

[0006] 3. There is no basis for classifying the voice data into more refined categories, which makes it inconvenient to extract the voice data later. Summary of the invention

[0007] The purpose of the present invention is to provide a voice data classification storage method and system based on an Internet of Things platform, which confirms the storage capacity of voice data to be stored according to the number of data segments, and can manage and access data more efficiently by classifying and integrating the data segments, thereby improving the efficiency of data processing and query, and selecting a storage channel whose remaining channel capacity is greater than the storage capacity of voice data to be stored as the first storage channel, thereby further improving the efficiency of storing and transmitting the voice data to be stored, and using a memory whose remaining storage capacity is greater than the storage capacity of voice data to be stored as the final storage for the voice data to be stored, thereby further ensuring the stability and durability of the voice data to be stored during storage, thereby solving the problems in the prior art.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The voice data classification storage method based on the Internet of Things platform includes the following steps:

[0010] S1: first obtain voice data from the IoT platform, perform data preprocessing on the obtained voice data, and mark the preprocessed voice data as voice data to be classified;

[0011] S2: performing category analysis on the speech data to be classified, classifying the speech data to be classified into categories according to different analysis categories, and marking the speech data to be classified after the attribute classification as speech data to be stored;

[0012] S3: confirming the data storage capacity of the voice data to be stored, and selecting a corresponding storage channel according to the data storage capacity of the voice data to be stored;

[0013] S4: After the storage channel selection of the voice data to be stored is completed, the voice data to be stored is stored in the corresponding memory, and each memory corresponds to a sub-memory of the same category as the voice data to be stored.

[0014] Preferably, data preprocessing is performed on the voice data acquired in S1, including:

[0015] The acquired voice data is sequentially subjected to data cleaning, signal pre-emphasis, data framing and windowing, voice enhancement and data normalization;

[0016] The voice data is first cleaned, and background noise, echo, and irrelevant audio tracks in the voice data are removed by using a filter, and the first voice data is obtained after the removal is completed;

[0017] Performing signal pre-emphasis on the first voice data, removing redundant information in the first voice data using a first-order difference method, and obtaining second voice data after the redundant information is removed;

[0018] Divide the second speech data into time frames and add a window function, wherein the window function is a Hamming window, and finally obtain the third speech data;

[0019] The third voice data is subjected to a time domain characteristic analysis by a time domain filtering method, wherein the time domain characteristic analysis is as follows: an output signal parameter of the third voice data is determined according to a unit impulse response of the third voice data on the filter, the output signal parameter is adjusted to be consistent with a preset output signal parameter, the preset output signal parameter is retrieved from a database, and the fourth voice data is obtained after the adjustment is completed;

[0020] The fourth speech data is normalized by cepstral mean normalization to finally obtain speech data to be classified.

[0021] Preferably, the first voice data is subjected to signal pre-emphasis, and redundant information in the first voice data is removed using a first-order difference method, and the second voice data is obtained after the redundant information is removed, including:

[0022] Determine a pre-emphasis coefficient corresponding to signal pre-emphasis of the first voice data;

[0023] The sampling time interval of the sampling points of the first voice data is obtained according to the pre-emphasis coefficient; wherein the sampling time interval is obtained by the following formula:

[0024]

[0025] Wherein, T represents the collection time interval of the sampling points of the first voice data; T0 represents the preset time interval reference value; λ represents the pre-emphasis coefficient; λ0 represents the preset pre-emphasis coefficient reference value, which is 0.972;

[0026] Extracting a sampling value corresponding to each sampling point in the first voice data according to the collection time interval;

[0027] The signal pre-emphasis model is used in combination with the pre-emphasis coefficient to obtain the pre-emphasized signal value corresponding to the sample value; wherein the signal pre-emphasis model is in the form of:

[0028] f(n)=x(n)-λ·x(n-1)

[0029] Wherein, f(n) represents the signal value after pre-emphasis processing corresponding to the sampling value; x(n) represents the current sampling value to be processed; x(n-1) represents the previous sampling value corresponding to the previous sampling value to be processed; λ represents the pre-emphasis coefficient;

[0030] Replacing the sampling value before pre-emphasis with the signal value after pre-emphasis to generate target voice data after signal pre-emphasis corresponding to the first voice data;

[0031] The redundant information in the first voice data is removed by using a first-order difference method for the target voice data, and the second voice data is obtained after the redundant information is removed.

[0032] Preferably, determining a pre-emphasis coefficient corresponding to signal pre-emphasis of the first voice data includes:

[0033] Extract frequency component information of a speech signal in the first speech data, and use the frequency component information of the speech signal to obtain a first coefficient parameter; wherein the first coefficient parameter is obtained by the following formula:

[0034]

[0035] Among them, ξ 01 represents the first coefficient parameter; m represents the number of speech frequency values ​​contained in the first speech data that exceed the preset speech frequency threshold; f i represents the i-th speech frequency value exceeding the preset speech frequency threshold; k represents the number of speech frequency values ​​contained in the first speech data that are lower than the preset speech frequency threshold; f j represents the jth speech frequency value that is lower than the preset speech frequency threshold; f0 represents the preset speech frequency threshold;

[0036] Extract the amplitude component information of the voice signal in the first voice data, and use the amplitude component information of the voice signal to obtain a second coefficient parameter; wherein the second coefficient parameter is obtained by the following formula:

[0037]

[0038] Among them, ξ 02 represents the first coefficient parameter; e represents the number of speech amplitude values ​​contained in the first speech data that exceed the preset speech amplitude threshold; F i represents the i-th speech amplitude value exceeding the preset speech amplitude threshold; r represents the number of speech amplitude values ​​contained in the first speech data that are lower than the preset speech amplitude threshold; F j represents the jth speech amplitude value that is lower than the preset speech amplitude threshold; F0 represents the preset speech amplitude threshold;

[0039] The pre-emphasis coefficient is obtained by using the first coefficient parameter and the second coefficient parameter; wherein the pre-emphasis coefficient is obtained by the following formula:

[0040]

[0041] Wherein, λ represents the pre-emphasis coefficient; λ0 represents the preset reference value of the emphasis coefficient, which is 0.972.

[0042] Preferably, the classification of the speech data to be classified in S2 includes:

[0043] Performing classification analysis on the voice data to be classified according to voice format, voice language, voice type, voice quality and voice content;

[0044] Among them, the category analysis is to integrate the voice data to be classified into voice format, voice language, voice type, voice quality and voice content;

[0045] The voice format is to integrate the voice data to be classified according to the formats of WAV, MP3, and MID I;

[0046] The speech language is to integrate the speech data to be classified according to the types of Chinese, English and French;

[0047] The voice type is to integrate the voice data to be classified according to structured data and unstructured data;

[0048] Voice quality is to integrate the voice data to be classified into high-quality voice and low-quality voice;

[0049] The speech content is to integrate the speech data to be classified according to semantics, sentence dictionary, and theme.

[0050] Preferably, the voice data to be classified is divided into categories and attributes according to different analysis categories in S2, including:

[0051] According to the type analysis of the speech data to be classified, the speech data to be classified is attributed according to the analyzed categories;

[0052] After the attribute division is completed, multiple sub-data of the attribute of the speech data to be classified are obtained, wherein the multiple sub-data are format attribute sub-data, type attribute sub-data, language type sub-data, quality type sub-data and content data sub-data;

[0053] And a plurality of sub-data of the attributes of the voice data to be classified are marked as voice data to be stored.

[0054] Preferably, the confirmation of the storage amount of the voice data to be stored in S3 includes:

[0055] Dividing the voice data to be stored into a plurality of data segments of equal length;

[0056] Performing data segment preprocessing on each data segment, wherein the data segment preprocessing is to remove abnormal values ​​and erroneous data from each data segment;

[0057] After the outliers and erroneous data are removed, the error range of the voice data to be stored is confirmed according to the error rules and the data category of the voice data to be stored, wherein the error rules are retrieved from the database;

[0058] Evaluate the data accuracy of the voice data to be stored according to the error range. If the data accuracy is not within the preset accuracy range, perform data segment preprocessing on the voice data to be stored again until the data accuracy is within the preset accuracy range.

[0059] After the data segment preprocessing is completed, the number of data segments is confirmed, and the length of the voice data to be stored is confirmed and acquired according to the number of data segments, and the storage capacity of the voice data to be stored is confirmed according to the length of the voice data to be stored.

[0060] Preferably, selecting a corresponding storage channel according to the data storage amount of the voice data to be stored in S3 includes:

[0061] The voice data to be stored is transferred to a storage channel of the memory for acquisition, wherein the number of the storage channels is not less than four;

[0062] Confirming the channel parameters of the stored channel, wherein the channel parameters are channel residual capacity and channel sensitivity;

[0063] Selecting a storage channel whose remaining channel capacity is greater than the amount of voice data to be stored as a first storage channel;

[0064] When the number of first storage channels is greater than one, selecting a channel with the strongest channel sensitivity among the plurality of first storage channels as a second storage channel;

[0065] Finally, the first storage channel or the second storage channel is transmitted to the final storage channel of the memory as the voice data to be stored.

[0066] The voice data classification storage system based on the Internet of Things platform includes:

[0067] Voice data storage module, used for:

[0068] Detecting the usage status of the memory, wherein the usage status of the memory is to detect whether the remaining storage capacity of the memory has reached a storage critical point. If the storage critical point has been reached, the memory is a memory that cannot be used as data storage;

[0069] Confirm the memory devices that can store data, wherein the number of the memory devices is not less than seven, and when the number of the memory devices is less than seven, automatically supplement the memory devices in the reserve storage space until the number of the memory devices is not less than seven;

[0070] Confirming the remaining storage capacity of each memory capable of storing data, confirming the storage location of the voice data to be stored according to the remaining storage capacity, and storing the voice data to be stored in the final memory through the final storage channel;

[0071] The remaining storage capacity of the memory is greater than the storage capacity of the voice data to be stored.

[0072] Preferably, it includes:

[0073] Category corresponding modules are used for:

[0074] Each memory has a plurality of sub-memories, wherein each sub-memory is a format sub-memory, a language sub-memory, a type sub-memory, a quality sub-memory and a content sub-memory;

[0075] Extracting attribute sub-data of the voice data to be stored in the memory;

[0076] After the attribute sub-data are extracted, the attribute sub-data are respectively mapped to the sub-memory of the same attribute according to the attribute category.

[0077] Compared with the prior art, the present invention has the following beneficial effects:

[0078] 1. The voice data classification storage method and system based on the Internet of Things platform provided by the present invention classifies the voice data to be classified according to voice format, voice language, voice type, voice quality and voice content, which can make data management more orderly and efficient. Different processing methods and management strategies can be adopted for different types of voice data, thereby simplifying the data management process.

[0079] 2. The voice data classification storage method and system based on the Internet of Things platform provided by the present invention confirms the storage capacity of the voice data to be stored according to the number of data segments. By classifying and integrating the data segments, the data can be managed and accessed more efficiently, thereby improving the efficiency of data processing and query. A storage channel whose remaining channel capacity is greater than the storage capacity of the voice data to be stored is selected as the first storage channel, thereby further improving the efficiency of storage and transmission of the voice data to be stored.

[0080] 3. The voice data classification storage method and system based on the Internet of Things platform provided by the present invention can further improve the security of the storage of the voice data to be stored according to the detection of the memory status, and use the memory whose remaining storage capacity is greater than the storage amount of the voice data to be stored as the final storage for the voice data to be stored, thereby further ensuring the stability and durability of the voice data to be stored during storage. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 A schematic diagram of the steps of classifying and storing voice data of the present invention;

[0082] Figure 2 The figure is a flow chart of voice data classification and storage according to the present invention. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0084] In order to solve the problem in the prior art that after acquiring the voice data, the voice data is not classified into more refined categories, which leads to inconvenience in extracting the voice data later, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0085] The voice data classification storage method based on the Internet of Things platform includes the following steps:

[0086] S1: first obtain voice data from the IoT platform, perform data preprocessing on the obtained voice data, and mark the preprocessed voice data as voice data to be classified;

[0087] Among them, data preprocessing of speech data can effectively reduce or eliminate background noise, spoken language differences and speech speed issues;

[0088] S2: performing category analysis on the speech data to be classified, classifying the speech data to be classified into categories according to different analysis categories, and marking the speech data to be classified after the attribute classification as speech data to be stored;

[0089] Among them, dividing the voice data to be classified according to category attributes can make data management more orderly and efficient;

[0090] S3: confirming the data storage capacity of the voice data to be stored, and selecting a corresponding storage channel according to the data storage capacity of the voice data to be stored;

[0091] Among them, a storage channel whose remaining channel capacity is greater than the storage amount of the voice data to be stored is selected as the first storage channel, which further improves the efficiency of the voice data to be stored when storing and transmitting, and avoids the problem of data congestion during the transmission process;

[0092] S4: After the storage channel for the voice data to be stored is selected, the voice data to be stored is stored in the corresponding memory, and each memory corresponds to a sub-memory of the same category as the voice data to be stored;

[0093] Among them, the memory whose remaining storage capacity is greater than the storage amount of the voice data to be stored is used as the final storage for the voice data to be stored, which further ensures the stability and durability of the voice data to be stored during storage.

[0094] The voice data to be acquired in S1 is preprocessed, including:

[0095] The acquired voice data is sequentially subjected to data cleaning, signal pre-emphasis, data framing and windowing, voice enhancement and data normalization;

[0096] The voice data is first cleaned, and background noise, echo, and irrelevant audio tracks in the voice data are removed by using a filter, and the first voice data is obtained after the removal is completed;

[0097] Performing signal pre-emphasis on the first voice data, removing redundant information in the first voice data using a first-order difference method, and obtaining second voice data after the redundant information is removed;

[0098] Divide the second speech data into time frames and add a window function, wherein the window function is a Hamming window, and finally obtain the third speech data;

[0099] The third voice data is subjected to a time domain characteristic analysis by a time domain filtering method, wherein the time domain characteristic analysis is as follows: an output signal parameter of the third voice data is determined according to a unit impulse response of the third voice data on the filter, the output signal parameter is adjusted to be consistent with a preset output signal parameter, the preset output signal parameter is retrieved from a database, and the fourth voice data is obtained after the adjustment is completed;

[0100] The fourth speech data is normalized by cepstral mean normalization to finally obtain speech data to be classified.

[0101] Specifically, the acquired voice data is first subjected to data cleaning, signal pre-emphasis, data framing and windowing, voice enhancement and data normalization in sequence. Data pre-processing of the voice data can effectively reduce or eliminate background noise, spoken language differences and speech speed problems, wherein data cleaning of the voice data can remove or correct erroneous, repeated or invalid voice data, thereby providing more accurate and reliable voice data, signal pre-emphasis of the voice data after data cleaning can improve the short-term stability of the voice signal, making the short-term voice signal more similar to a periodic function, thereby facilitating Fourier transform, data framing and windowing of the voice data after signal pre-emphasis can improve the frequency domain resolution and time domain resolution of the voice signal, and improve the accuracy and efficiency of voice signal processing, voice enhancement after data framing and windowing can effectively remove background noise, make the voice signal clearer, and improve the accuracy of voice recognition, data normalization after voice enhancement, first subtracting the mean of the MFCC features in the voice data, can reduce the impact of acoustic channel distortion. The mean of the i-th dimension of all frame features of the speech data is estimated, where T represents the number of i-th dimension feature frames in the sentence, and then the mean is subtracted from all i-th dimension feature frames in the sentence, thereby eliminating noise and irregularities in the speech data, such as capitalization differences, spelling errors, synonyms, etc., thereby improving the accuracy of text analysis.

[0102] Specifically, the first voice data is subjected to signal pre-emphasis, and redundant information in the first voice data is removed by using a first-order difference method, and the second voice data is obtained after the redundant information is removed, including:

[0103] Determine a pre-emphasis coefficient corresponding to signal pre-emphasis of the first voice data;

[0104] The sampling time interval of the sampling points of the first voice data is obtained according to the pre-emphasis coefficient; wherein the sampling time interval is obtained by the following formula:

[0105]

[0106] Wherein, T represents the collection time interval of the sampling points of the first voice data; T0 represents the preset time interval reference value; λ represents the pre-emphasis coefficient; λ0 represents the preset pre-emphasis coefficient reference value, which is 0.972;

[0107] Extracting a sampling value corresponding to each sampling point in the first voice data according to the collection time interval;

[0108] The signal pre-emphasis model is used in combination with the pre-emphasis coefficient to obtain the pre-emphasized signal value corresponding to the sample value; wherein the signal pre-emphasis model is in the form of:

[0109] f(n)=x(n)-λ·x(n-1)

[0110] Wherein, f(n) represents the signal value after pre-emphasis processing corresponding to the sampling value; x(n) represents the current sampling value to be processed; x(n-1) represents the previous sampling value corresponding to the previous sampling value to be processed; λ represents the pre-emphasis coefficient;

[0111] Replacing the sampling value before pre-emphasis with the signal value after pre-emphasis to generate target voice data after signal pre-emphasis corresponding to the first voice data;

[0112] The redundant information in the first voice data is removed by using a first-order difference method for the target voice data, and the second voice data is obtained after the redundant information is removed.

[0113] The technical effect of the above technical solution is: by performing signal pre-emphasis processing on the first voice data, the high-frequency part of the voice signal can be enhanced, making the spectrum of the voice signal flatter. This helps to improve the clarity and intelligibility of the voice signal, especially the information in the high-frequency part is easier to identify and distinguish. Dynamically adjusting the acquisition time interval of the sampling point according to the pre-emphasis coefficient can ensure that the sampling process is more reasonable and effective under different pre-emphasis levels. This adaptive adjustment capability makes the sampling process more flexible and can adapt to the characteristics of different voice signals.

[0114] The sample value of each sampling point is extracted according to the adjusted acquisition time interval, and the sample value is accurately processed using the signal pre-emphasis model. This ensures that important information of the original speech signal is not lost or distorted during the processing, thereby maintaining the integrity and accuracy of the speech signal. By replacing the sample value before pre-emphasis with the signal value after pre-emphasis, optimized target speech data can be generated. This reduces the amount of calculation and complexity in the subsequent speech processing process, and improves the efficiency and performance of speech processing.

[0115] By applying the first-order difference method to the target speech data, redundant information in the speech data can be removed, and the speech signal can be further compressed and optimized. This helps to reduce the overhead of data storage and transmission while maintaining the key features and recognizability of the speech signal.

[0116] Specifically, determining a pre-emphasis coefficient corresponding to signal pre-emphasis of the first voice data includes:

[0117] Extract frequency component information of a speech signal in the first speech data, and use the frequency component information of the speech signal to obtain a first coefficient parameter; wherein the first coefficient parameter is obtained by the following formula:

[0118]

[0119] Among them, ξ 01represents the first coefficient parameter; m represents the number of speech frequency values ​​contained in the first speech data that exceed the preset speech frequency threshold; f i represents the i-th speech frequency value exceeding the preset speech frequency threshold; k represents the number of speech frequency values ​​contained in the first speech data that are lower than the preset speech frequency threshold; f j represents the jth speech frequency value that is lower than the preset speech frequency threshold; f0 represents the preset speech frequency threshold;

[0120] Extract the amplitude component information of the voice signal in the first voice data, and use the amplitude component information of the voice signal to obtain a second coefficient parameter; wherein the second coefficient parameter is obtained by the following formula:

[0121]

[0122] Among them, ξ 02 represents the first coefficient parameter; e represents the number of speech amplitude values ​​contained in the first speech data that exceed the preset speech amplitude threshold; F i represents the i-th speech amplitude value exceeding the preset speech amplitude threshold; r represents the number of speech amplitude values ​​contained in the first speech data that are lower than the preset speech amplitude threshold; F j represents the jth speech amplitude value that is lower than the preset speech amplitude threshold; F0 represents the preset speech amplitude threshold;

[0123] The pre-emphasis coefficient is obtained by using the first coefficient parameter and the second coefficient parameter; wherein the pre-emphasis coefficient is obtained by the following formula:

[0124]

[0125] Wherein, λ represents the pre-emphasis coefficient; λ0 represents the preset reference value of the emphasis coefficient, which is 0.972.

[0126] The technical effect of the above technical solution is that by extracting the frequency component information of the voice signal in the first voice data, the distribution of the voice signal in different frequency bands can be understood, which helps to identify the high-frequency and low-frequency components in the voice signal and provides a basis for subsequent pre-emphasis processing.

[0127] When analyzing frequency components, the preset voice frequency threshold is used for screening, and low-frequency components below the threshold or high-frequency components above the threshold can be excluded. This adaptive threshold setting method can be flexibly adjusted according to the characteristics of the voice signal, thereby improving the accuracy and adaptability of the analysis. At the same time, by extracting the amplitude component information of the voice signal in the first voice data, the distribution of the voice signal in different amplitude segments can be further understood. This helps to identify the strong and weak sound components in the voice signal, and provides more reference information for subsequent pre-emphasis processing. Similarly, by screening using the preset voice amplitude threshold, weak sound components below the threshold or strong sound components above the threshold can be excluded. This adaptive threshold setting method can be flexibly adjusted according to the characteristics of the voice signal, thereby further improving the accuracy and adaptability of the analysis.

[0128] When obtaining the first coefficient parameter and the second coefficient parameter, the frequency component and amplitude component information of the speech signal are considered at the same time. This comprehensive analysis method can more comprehensively understand the characteristics of the speech signal and provide a more accurate basis for determining the pre-emphasis coefficient. According to the first coefficient parameter and the second coefficient parameter, the pre-emphasis coefficient is calculated by a specific formula. This dynamic adjustment method can be flexibly adjusted according to the actual characteristics of the speech signal to ensure that the effect of the pre-emphasis processing is more in line with actual needs. The pre-emphasis coefficient determined by the above technical solution can more accurately reflect the characteristics of the speech signal, thereby optimizing the effect of the pre-emphasis processing. This helps to improve the quality and intelligibility of the speech signal and provide a better foundation for subsequent speech processing tasks.

[0129] In summary, the above technical solution can determine the pre-emphasis coefficient by comprehensively considering the frequency and amplitude component information of the speech signal, thereby optimizing the effect of speech signal processing.

[0130] The voice data to be classified in S2 is analyzed by category, including:

[0131] Performing classification analysis on the voice data to be classified according to voice format, voice language, voice type, voice quality and voice content;

[0132] Among them, the category analysis is to integrate the voice data to be classified into voice format, voice language, voice type, voice quality and voice content;

[0133] The voice format is to integrate the voice data to be classified according to the formats of WAV, MP3, and MID I;

[0134] The speech language is to integrate the speech data to be classified according to the types of Chinese, English and French;

[0135] The voice type is to integrate the voice data to be classified according to structured data and unstructured data;

[0136] Voice quality is to integrate the voice data to be classified into high-quality voice and low-quality voice;

[0137] The speech content is to integrate the speech data to be classified according to semantics, sentence dictionary, and theme.

[0138] According to different analysis categories in S2, the voice data to be classified is divided into categories and attributes, including:

[0139] According to the type analysis of the speech data to be classified, the speech data to be classified is attributed according to the analyzed categories;

[0140] After the attribute division is completed, multiple sub-data of the attribute of the speech data to be classified are obtained, wherein the multiple sub-data are format attribute sub-data, type attribute sub-data, language type sub-data, quality type sub-data and content data sub-data;

[0141] And a plurality of sub-data of the attributes of the voice data to be classified are marked as voice data to be stored.

[0142] Specifically, the voice data is first analyzed by category according to the voice format, voice language, voice type, voice quality and voice content. The category analysis is to extract and analyze each type of the voice data to be classified. Among them, the voice format analysis of the voice data to be classified can ensure that the voice data is compatible with specific equipment or software, avoiding the problem of being unable to play or process due to format mismatch. The voice language analysis can better analyze and understand the voice characteristics of different languages, so as to effectively process and apply the characteristics of different languages. The voice type analysis can better identify the voice commands, voice chats, voice messages, etc. in the voice data to be classified. When performing quality analysis, different voice data qualities will affect its subsequent processing and analysis. High-quality voice data can provide more accurate analysis results, while low-quality voice data may cause errors. Classifying voice data by quality can facilitate subsequent data processing and analysis. When performing content analysis, the content of voice data contains a lot of semantic information. By classifying the speech content, we can better understand the speaker's intention and information, so as to conduct more accurate analysis and processing. Then, we can divide the attribute data of the speech data to be classified according to the category analysis, and obtain the format attribute sub-data, type attribute sub-data, language type sub-data, quality type sub-data and content data sub-data after the division, so as to accurately obtain the classification data of different categories in the speech data, which will also facilitate the subsequent data retrieval. The speech data to be classified can be classified according to the speech format, speech language, speech type, speech quality and speech content, which can make the data management more orderly and efficient. Different processing methods and management strategies can be adopted for different types of speech data, so as to simplify the data management process.

[0143] In order to solve the problem that in the prior art, effective data transmission is not performed according to the storage amount of voice data, resulting in traffic congestion during data transmission to the storage, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0144] Confirmation of the storage capacity of voice data to be stored in S3, including:

[0145] Dividing the voice data to be stored into a plurality of data segments of equal length;

[0146] Performing data segment preprocessing on each data segment, wherein the data segment preprocessing is to remove abnormal values ​​and erroneous data from each data segment;

[0147] After the outliers and erroneous data are removed, the error range of the voice data to be stored is confirmed according to the error rules and the data category of the voice data to be stored, wherein the error rules are retrieved from the database;

[0148] Evaluate the data accuracy of the voice data to be stored according to the error range. If the data accuracy is not within the preset accuracy range, perform data segment preprocessing on the voice data to be stored again until the data accuracy is within the preset accuracy range.

[0149] After the data segment preprocessing is completed, the number of data segments is confirmed, and the length of the voice data to be stored is confirmed and acquired according to the number of data segments, and the storage capacity of the voice data to be stored is confirmed according to the length of the voice data to be stored.

[0150] Selecting a corresponding storage channel according to the data storage amount of the voice data to be stored in S3 includes:

[0151] The voice data to be stored is transferred to a storage channel of the memory for acquisition, wherein the number of the storage channels is not less than four;

[0152] Confirming the channel parameters of the stored channel, wherein the channel parameters are channel residual capacity and channel sensitivity;

[0153] Selecting a storage channel whose remaining channel capacity is greater than the amount of voice data to be stored as a first storage channel;

[0154] When the number of first storage channels is greater than one, selecting a channel with the strongest channel sensitivity among the plurality of first storage channels as a second storage channel;

[0155] Finally, the first storage channel or the second storage channel is transmitted to the final storage channel of the memory as the voice data to be stored.

[0156] Specifically, preprocessing the data segments of the voice data to be stored can remove noise and outliers in the data, thereby improving the accuracy and reliability of the data. Data from different data sources can be integrated to eliminate data redundancy and duplication, and improve data consistency and integrity. The storage capacity of the voice data to be stored is then confirmed based on the number of data segments. By classifying and integrating the data segments, data can be managed and accessed more efficiently, the efficiency of data processing and query can be improved, data security policies can be implemented more conveniently, data integration and analysis can be achieved more easily, the associations and rules between data can be discovered, and more powerful support can be provided for decision-making. The channel remaining capacity and channel sensitivity of the storage channel are then confirmed, and a storage channel whose channel remaining capacity is greater than the storage capacity of the voice data to be stored is selected as the first storage channel, which further improves the efficiency of the storage and transmission of the voice data to be stored and avoids the problem of data congestion during the transmission process. The channel with the strongest channel sensitivity among multiple first storage channels is selected as the second storage channel, which further improves the transmission speed of the voice data to be stored.

[0157] In order to solve the problem in the prior art that when voice data is stored, the corresponding storage space is not selected according to the size of the voice data, and the voice data is not stored more finely according to the category attributes, thereby reducing the storage and later retrieval efficiency, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0158] The voice data classification storage system based on the Internet of Things platform includes:

[0159] Voice data storage module, used for:

[0160] Detecting the usage status of the memory, wherein the usage status of the memory is to detect whether the remaining storage capacity of the memory has reached a storage critical point. If the storage critical point has been reached, the memory is a memory that cannot be used as data storage;

[0161] Confirm the memory devices that can store data, wherein the number of the memory devices is not less than seven, and when the number of the memory devices is less than seven, automatically supplement the memory devices in the reserve storage space until the number of the memory devices is not less than seven;

[0162] Confirming the remaining storage capacity of each memory capable of storing data, confirming the storage location of the voice data to be stored according to the remaining storage capacity, and storing the voice data to be stored in the final memory through the final storage channel;

[0163] The remaining storage capacity of the memory is greater than the storage capacity of the voice data to be stored.

[0164] Category corresponding modules are used for:

[0165] Each memory has a plurality of sub-memories, wherein each sub-memory is a format sub-memory, a language sub-memory, a type sub-memory, a quality sub-memory and a content sub-memory;

[0166] Extracting attribute sub-data of the voice data to be stored in the memory;

[0167] After the attribute sub-data are extracted, the attribute sub-data are respectively mapped to the sub-memory of the same attribute according to the attribute category.

[0168] Specifically, the storage location of the voice data to be stored is confirmed by the voice data storage module, wherein the state of each storage is detected before the voice data to be stored is stored, and the security of the storage of the voice data to be stored can be further improved according to the detection of the storage state, and the storage whose remaining storage capacity is greater than the storage amount of the voice data to be stored is used as the final storage of the voice data to be stored, which further ensures the stability and durability of the voice data to be stored during storage, and when the storage location of the voice data to be stored is confirmed, the voice data to be stored in the storage is stored in the format sub-memory, language sub-memory, type sub-memory, quality sub-memory and content sub-memory according to the attribute category of the attribute sub-data, which further improves the integrity of the voice data classification and facilitates the subsequent retrieval efficiency.

[0169] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0170] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A voice data classification and storage method based on the Internet of Things platform, characterized in that: The steps include: S1: first obtain voice data from the IoT platform, perform data preprocessing on the obtained voice data, and mark the preprocessed voice data as voice data to be classified; S2: performing category analysis on the speech data to be classified, classifying the speech data to be classified into categories according to different analysis categories, and marking the speech data to be classified after the attribute classification as speech data to be stored; S3: confirming the data storage capacity of the voice data to be stored, and selecting a corresponding storage channel according to the data storage capacity of the voice data to be stored; S4: After the storage channel for the voice data to be stored is selected, the voice data to be stored is stored in the corresponding memory, and each memory corresponds to a sub-memory of the same category as the voice data to be stored; The voice data to be acquired in S1 is preprocessed, including: The acquired voice data is sequentially subjected to data cleaning, signal pre-emphasis, data framing and windowing, voice enhancement and data normalization; The voice data is first cleaned, and background noise, echo, and irrelevant audio tracks in the voice data are removed by using a filter, and the first voice data is obtained after the removal is completed; Performing signal pre-emphasis on the first voice data, removing redundant information in the first voice data using a first-order difference method, and obtaining second voice data after the redundant information is removed; Divide the second speech data into time frames and add a window function, wherein the window function is a Hamming window, and finally obtain the third speech data; The third voice data is subjected to a time domain characteristic analysis by a time domain filtering method, wherein the time domain characteristic analysis is as follows: an output signal parameter of the third voice data is determined according to a unit impulse response of the third voice data on the filter, the output signal parameter is adjusted to be consistent with a preset output signal parameter, the preset output signal parameter is retrieved from a database, and the fourth voice data is obtained after the adjustment is completed; Normalizing the fourth speech data by cepstrum mean normalization to finally obtain speech data to be classified; The first voice data is subjected to signal pre-emphasis, and redundant information in the first voice data is removed by using a first-order difference method, and second voice data is obtained after the redundant information is removed, including: Determine a pre-emphasis coefficient corresponding to signal pre-emphasis of the first voice data; The sampling time interval of the sampling points of the first voice data is obtained according to the pre-emphasis coefficient; wherein the sampling time interval is obtained by the following formula: Wherein, T represents the sampling time interval of the sampling points of the first speech data; T0 represents the preset time interval reference value; λ represents the pre-emphasis coefficient; λ0 represents the preset pre-emphasis coefficient reference value, which is 0.972; Extracting a sampling value corresponding to each sampling point in the first voice data according to the sampling time interval; The signal pre-emphasis model is used in combination with the pre-emphasis coefficient to obtain the pre-emphasized signal value corresponding to the sample value; wherein the signal pre-emphasis model is in the form of: Among them, f(n) represents the signal value after pre-emphasis processing corresponding to the sampling value; x(n) represents the current sampling value to be processed; x(n-1) represents the previous sampling value corresponding to the previous sampling value to be processed; λ represents the pre-emphasis coefficient; Replacing the sampling value before pre-emphasis with the signal value after pre-emphasis to generate target voice data after signal pre-emphasis corresponding to the first voice data; Using a first-order difference method to remove redundant information in the first voice data, the target voice data is used to obtain second voice data after the redundant information is removed; Determining a pre-emphasis coefficient corresponding to signal pre-emphasis of the first voice data includes: Extract frequency component information of a speech signal in the first speech data, and use the frequency component information of the speech signal to obtain a first coefficient parameter; wherein the first coefficient parameter is obtained by the following formula: Among them, ξ 01 represents the first coefficient parameter; m represents the number of speech frequency values ​​contained in the first speech data that exceed the preset speech frequency threshold; f i represents the i-th speech frequency value exceeding the preset speech frequency threshold; k represents the number of speech frequency values ​​contained in the first speech data that are lower than the preset speech frequency threshold; f j represents the jth speech frequency value that is lower than the preset speech frequency threshold; f0 represents the preset speech frequency threshold; Extract the amplitude component information of the voice signal in the first voice data, and use the amplitude component information of the voice signal to obtain a second coefficient parameter; wherein the second coefficient parameter is obtained by the following formula: Among them, ξ 02 represents the first coefficient parameter; e represents the number of speech amplitude values ​​contained in the first speech data that exceed the preset speech amplitude threshold; F i represents the i-th speech amplitude value exceeding the preset speech amplitude threshold; r represents the number of speech amplitude values ​​contained in the first speech data that are lower than the preset speech amplitude threshold; F j represents the jth speech amplitude value that is lower than the preset speech amplitude threshold; F0 represents the preset speech amplitude threshold; The pre-emphasis coefficient is obtained by using the first coefficient parameter and the second coefficient parameter; wherein the pre-emphasis coefficient is obtained by the following formula: Wherein, λ represents the pre-emphasis coefficient; λ0 represents the preset reference value of the emphasis coefficient, which is 0.972; Confirmation of the storage capacity of voice data to be stored in S3, including: Dividing the voice data to be stored into a plurality of data segments of equal length; Performing data segment preprocessing on each data segment, wherein the data segment preprocessing is to remove abnormal values ​​and erroneous data from each data segment; After the outliers and erroneous data are removed, the error range of the voice data to be stored is confirmed according to the error rules and the data category of the voice data to be stored, wherein the error rules are retrieved from the database; Evaluate the data accuracy of the voice data to be stored according to the error range. If the data accuracy is not within the preset accuracy range, perform data segment preprocessing on the voice data to be stored again until the data accuracy is within the preset accuracy range. After the data segment preprocessing is completed, the number of data segments is confirmed, and the length of the voice data to be stored is confirmed according to the number of data segments for acquisition, and the storage capacity of the voice data to be stored is confirmed according to the length of the voice data to be stored; Selecting a corresponding storage channel according to the data storage amount of the voice data to be stored in S3 includes: The voice data to be stored is transferred to a storage channel of the memory for acquisition, wherein the number of the storage channels is not less than four; Confirming the channel parameters of the stored channel, wherein the channel parameters are channel residual capacity and channel sensitivity; Selecting a storage channel whose remaining channel capacity is greater than the amount of voice data to be stored as a first storage channel; When the number of first storage channels is greater than one, selecting a channel with the strongest channel sensitivity among the plurality of first storage channels as a second storage channel; Finally, the first storage channel or the second storage channel is transmitted to the final storage channel of the memory as the voice data to be stored.

2. The method for classifying and storing voice data based on the Internet of Things platform according to claim 1 is characterized in that: The voice data to be classified in S2 is analyzed by category, including: Performing classification analysis on the voice data to be classified according to voice format, voice language, voice type, voice quality and voice content; Among them, the category analysis is to integrate the voice data to be classified into voice format, voice language, voice type, voice quality and voice content; The voice format is to integrate the voice data to be classified according to the formats of WAV, MP3 and MIDI; The speech language is to integrate the speech data to be classified according to the types of Chinese, English and French; The voice type is to integrate the voice data to be classified according to structured data and unstructured data; Voice quality is to integrate the voice data to be classified into high-quality voice and low-quality voice; The speech content is to integrate the speech data to be classified according to semantics, sentence dictionary, and theme.

3. The method for classifying and storing voice data based on the Internet of Things platform according to claim 2 is characterized in that: According to different analysis categories in S2, the voice data to be classified is divided into categories and attributes, including: According to the type analysis of the speech data to be classified, the speech data to be classified is attributed according to the analyzed categories; After the attribute division is completed, multiple sub-data of the attribute of the speech data to be classified are obtained, wherein the multiple sub-data are format attribute sub-data, type attribute sub-data, language type sub-data, quality type sub-data and content data sub-data; And a plurality of sub-data of the attributes of the voice data to be classified are marked as voice data to be stored.

4. A voice data classification storage system based on an Internet of Things platform, applied in the voice data classification storage method based on an Internet of Things platform according to claim 3, comprising: Voice data storage module, used for: Detecting the usage status of the memory, wherein the usage status of the memory is to detect whether the remaining storage capacity of the memory has reached a storage critical point. If the storage critical point has been reached, the memory is a memory that cannot be used as data storage; Confirm the memory devices that can store data, wherein the number of the memory devices is not less than seven, and when the number of the memory devices is less than seven, automatically supplement the memory devices in the reserve storage space until the number of the memory devices is not less than seven; Confirming the remaining storage capacity of each memory capable of storing data, confirming the storage location of the voice data to be stored according to the remaining storage capacity, and storing the voice data to be stored in the final memory through the final storage channel; The remaining storage capacity of the memory is greater than the storage capacity of the voice data to be stored.

5. The voice data classification storage system based on the Internet of Things platform according to claim 4 is characterized in that: include: Category corresponding modules are used for: Each memory has a plurality of sub-memories, wherein each sub-memory is a format sub-memory, a language sub-memory, a type sub-memory, a quality sub-memory and a content sub-memory; Extracting attribute sub-data of the voice data to be stored in the memory; After the attribute sub-data are extracted, the attribute sub-data are respectively mapped to the sub-memory of the same attribute according to the attribute category.