Remote voice control method and system for numerical control system

Abstract

The present invention relates to the field of speech analysis technology, in particular to a remote speech control method and system for a numerical control system, the system comprising a feature extraction module, a noise classification and detection module and a time window scale determination module, wherein: the feature extraction module acquires a noise signal and extracts a spectrum feature and a time feature of the noise signal; the noise classification and detection module classifies the noise signal according to the spectrum feature and the time feature of the noise signal, and generates different types of noise signals; according to different types of noise signals, different time window scales are selected to perform noise detection on the noise signal, wherein: the time window scale determination module calculates the sizes of different time window scales according to the spectrum feature and the time feature of the noise signal, wherein the different time window scales include a long time window scale and a short time window scale.

Description

Technical Field

[0001] The present invention relates to the technical field of speech analysis, and in particular to a remote speech control method and system for a numerical control system. Background Art

[0002] In traditional CNC system remote voice control, noise detection and gain control are often processed using a fixed window size.

[0003] This method performs poorly in different types of noise environments and has a variety of significant problems. First, for stable noise (such as mechanical noise, environmental noise), a fixed time window scale may not be able to fully capture its long-term characteristics, resulting in insufficient stability in noise detection. For example, machine tool operation noise is usually relatively uniform and has stable spectral characteristics, but when using a shorter time window for detection, the fluctuation of instantaneous noise may be mistaken for real noise, causing false detection and missed detection.

[0004] Secondly, for instantaneous noise (such as wind noise and burst noise), a fixed time window scale may not be able to capture its instantaneous changes in time, resulting in insufficient real-time performance. Wind noise and burst noise have high frequency change rates and instantaneous characteristics. If a longer time window is used for detection, the important features of these instantaneous noises may be missed, making it impossible for the system to adjust the gain control strategy in time.

[0005] In addition, the traditional fixed window size method often cannot be optimized according to different noise types when processing noise signals, which not only wastes computing resources but also reduces the overall efficiency and response speed of the system. For example, in a stable mechanical noise environment, using a shorter time window will increase the computational burden, while in an instantaneous wind noise environment, using a longer time window will cause processing delays.

[0006] The above problems have seriously affected the performance and user experience of remote voice control of CNC systems. Summary of the invention

[0007] The purpose of the present invention is to provide a remote voice control method and system for a numerical control system to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, one of the objectives of the present invention is to provide a remote voice control system for a numerical control system, comprising a feature extraction module, a noise classification and detection module and a time window scale determination module, wherein:

[0009] The feature extraction module acquires the noise signal and extracts the frequency spectrum feature and time feature of the noise signal;

[0010] The noise classification and detection module classifies the noise signal according to the spectral characteristics and time characteristics of the noise signal to generate different types of noise signals; according to different types of noise signals, different time window scales are selected to perform noise detection on the noise signal, wherein:

[0011] The time window scale determination module calculates the sizes of different time window scales according to the frequency spectrum characteristics and time characteristics of the noise signal, wherein the different time window scales include a long time window scale and a short time window scale.

[0012] As a further improvement of the technical solution, the feature extraction module uses a multi-microphone array to collect noise signals in the numerical control system environment.

[0013] As a further improvement of the present technical solution, the feature extraction module divides the collected noise signal into multiple fixed time windows, performs Fourier transform on the noise signal in each fixed time window, obtains the spectrum amplitude in the time window, and calculates the spectrum energy distribution of the spectrum amplitude in different frequency bands, the frequency bands including low frequency band, mid frequency band and high frequency band; and uses the spectrum amplitude and spectrum energy distribution as spectrum features.

[0014] As a further improvement of the present technical solution, the process of extracting time features by the feature extraction module is as follows:

[0015] In each fixed time window, calculate the average energy of the noise signal , the formula is as follows:

[0016] ,in is the number of sampling points in a fixed time window, It is The noise signal amplitude of each sampling point;

[0017] In each fixed time window, calculate the energy variance of the noise signal , the formula is as follows:

[0018] ;

[0019] In each fixed time window, calculate the zero crossing rate of the noise signal , the formula is as follows:

[0020] ,in is a sign function;

[0021] The average energy , Energy variance and zero crossing rate As a time feature.

[0022] As a further improvement of the technical solution, the noise classification and detection module includes a noise classification unit, and the process of the noise classification unit generating different types of noise signals specifically includes:

[0023] The extracted spectral features and time features are combined into a feature vector as a fused feature vector;

[0024] Different types of noise signal samples are collected from public databases and used as training data for the classification model. The sample types include stable noise and transient noise.

[0025] According to the classification model structure, the classification model is trained using the training data to generate a trained noise signal classification model;

[0026] The trained noise signal classification model is used to classify the fused feature vector to generate different types of noise signals.

[0027] As a further improvement of the present technical solution, the noise classification unit selects a convolutional neural network as a classification model for noise signals, and designs a structure of the classification model including an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer, wherein the input layer receives training data; each convolution layer contains multiple convolution kernels for extracting local features of the training data; the pooling layer is used to reduce the feature dimension and reduce the amount of calculation; the fully connected layer is used to map the extracted features to different noise categories; and the output layer outputs the category of the noise signal.

[0028] As a further improvement of the technical solution, the noise classification and detection module includes a noise detection unit, which selects different time window scales according to different types of noise signals to perform noise detection on the noise signal, specifically including:

[0029] For the noise signal of the stable noise type, a long time window scale is used as the time scale window for noise detection;

[0030] For noise signals of the transient noise type, a short time window scale is used as the time scale window for noise detection.

[0031] As a further improvement of the technical solution, the process of determining the size of the short time window scale in the time window scale determination module specifically includes:

[0032] According to the average energy in the temporal characteristics of the noise signal and zero crossing rate , calculate the size of the short time window scale, the calculation formula is as follows:

[0033] ,in Indicates the size of the short time window; and Represents the average energy and zero crossing rate The weight coefficient is determined by the optimization algorithm; A positive number to avoid zero denominator.

[0034] As a further improvement of the technical solution, the process of determining the size of the long time window scale in the time window scale determination module specifically includes:

[0035] According to the energy variance in the temporal characteristics of the noise signal The band energy distribution of the spectral characteristics is used to calculate the long-term window size. The calculation formula is as follows:

[0036] ,in Indicates the size of the long time window; and Represents the energy variance and weight coefficients of frequency band energy distribution of spectral characteristics; A positive number to avoid The function's argument is zero; Indicates the low-frequency band energy given in the frequency band energy distribution; Indicates the high-frequency band energy in the frequency band energy distribution.

[0037] A second object of the present invention is to provide a method for a remote voice control system of a numerical control system, comprising the following method steps:

[0038] S1. Obtain a noise signal and extract the spectral characteristics and time characteristics of the noise signal;

[0039] S2. Classify the noise signal according to the spectral characteristics and time characteristics of the noise signal to generate different types of noise signals; select different time window scales according to different types of noise signals to perform noise detection on the noise signal, wherein:

[0040] According to the frequency spectrum characteristics and time characteristics of the noise signal, the sizes of different time window scales are calculated, wherein the different time window scales include a long time window scale and a short time window scale.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The remote voice control method and system of a numerical control system extracts the spectrum characteristics and time characteristics of the noise signal and uses a convolutional neural network to classify the noise type; for the identified stable noise, a long time window scale is selected to fully capture its long-term characteristics and reduce the influence of instantaneous fluctuations, thereby improving the stability and accuracy of noise detection and reducing false detection and missed detection; by extracting the spectrum characteristics and time characteristics and using a convolutional neural network for classification; for the identified instantaneous noise, the noise detection unit selects a short time window scale to quickly capture its instantaneous changes and improve the real-time performance of the detection;

[0043] In addition, the window size is dynamically calculated according to the characteristics of the noise signal to ensure that the gain control strategy can be adjusted in time in different noise environments, optimize resource utilization, and improve response speed and overall efficiency; the noise signal is classified to identify stable noise and transient noise; the time window size required for different types of noise is dynamically calculated and selected according to the classification results, avoiding the problem of increasing the computational burden by using short time windows in stable noise environments and causing processing delays by using long time windows in transient noise environments, thereby optimizing resource utilization and improving the overall efficiency and response speed of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0045] Figure 2 It is a schematic diagram of the overall module of the present invention.

[0046] In the figure: 100, feature extraction module; 200, noise classification and detection module; 201, noise classification unit; 202, noise detection unit; 300, time window scale determination module. DETAILED DESCRIPTION

[0047] The following will be combined with the accompanying drawings in 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.

[0048] Next, see Figure 1-Figure 2 One of the purposes of this embodiment is to provide a remote voice control system for a numerical control system, including a feature extraction module 100, a noise classification and detection module 200 and a time window scale determination module 300.

[0049] The feature extraction module 100 uses a multi-microphone array to collect noise signals in the CNC system environment. The multi-microphone array can improve the signal collection quality and the extraction accuracy of time-frequency features; the collected noise signals are preprocessed, including filtering, normalization and other steps to remove unnecessary interference and improve signal quality.

[0050] The feature extraction module 100 extracts the spectrum features and time features of the preprocessed noise signal, specifically including:

[0051] The collected noise signal is divided into multiple fixed time windows, which helps to capture the spectral characteristics of the signal at different time points. The noise signal in each fixed time window is Fourier transformed to obtain the spectral amplitude in the time window, and the spectral energy distribution of the spectral amplitude in different frequency bands is calculated. The frequency bands include low frequency band (0-500Hz), medium frequency band (500-2000Hz) and high frequency band (2000-8000Hz); the spectral amplitude and spectral energy distribution are used as spectral features;

[0052] In each fixed time window, calculate the average energy of the noise signal , the formula is as follows:

[0053] ,in is the number of sampling points in a fixed time window, It is The noise signal amplitude of each sampling point;

[0054] In each fixed time window, calculate the energy variance of the noise signal , the formula is as follows:

[0055] ;

[0056] In each fixed time window, calculate the zero crossing rate of the noise signal , the formula is as follows:

[0057] ,in is a sign function;

[0058] The average energy , Energy variance and zero crossing rate As a time feature.

[0059] The noise classification unit 201 in the noise classification and detection module 200 classifies the noise signal according to the spectral characteristics and time characteristics of the noise signal, and generates different types of noise signals, specifically including:

[0060] The extracted spectral features and time features are combined into a feature vector as a fusion feature vector, and a convolutional neural network is selected as a classification model for noise signals;

[0061] Different types of noise signal samples are collected from public databases and used as training data for the classification model. The sample types include stable noise and transient noise.

[0062] Design a classification model structure, including input layer, convolution layer, pooling layer, fully connected layer and output layer, where the input layer receives training data; each convolution layer contains multiple convolution kernels to extract local features of training data; the pooling layer is used to reduce feature dimensions and reduce the amount of calculation; the fully connected layer is used to map the extracted features to different noise categories; the output layer outputs the category of the noise signal;

[0063] According to the designed classification model structure, the classification model is trained using the training data to generate a trained noise signal classification model;

[0064] The trained noise signal classification model is used to classify the fused feature vector to generate different types of noise signals.

[0065] The noise detection unit 202 in the noise classification and detection module 200 selects different time window scales according to different types of noise signals to perform noise detection on the noise signal, wherein:

[0066] The time window scale determination module 300 calculates the sizes of different time window scales according to the spectral characteristics and time characteristics of the noise signal, wherein the different time window scales include a long time window scale and a short time window scale, specifically including:

[0067] According to the average energy in the temporal characteristics of the noise signal and zero crossing rate , calculate the size of the short time window scale, the calculation formula is as follows:

[0068] ,in Indicates the size of the short time window; and Represents the average energy and zero crossing rate The weight coefficient is determined by the optimization algorithm; is a positive number to avoid the denominator being zero;

[0069] According to the energy variance in the temporal characteristics of the noise signal The band energy distribution of the spectral characteristics is used to calculate the long-term window size. The calculation formula is as follows:

[0070] ,in Indicates the size of the long time window; and Represents the energy variance and weight coefficients of frequency band energy distribution of spectral characteristics; A positive number to avoid The function's argument is zero; Indicates the low-frequency band energy given in the frequency band energy distribution; Indicates the high-frequency band energy in the frequency band energy distribution.

[0071] The noise detection unit 202 selects different time window scales according to different types of noise signals to perform noise detection on the noise signal, specifically including:

[0072] For noise signals of stable noise type, a long time window scale is used as the time scale window for noise detection; stable noise is usually more uniform, and the use of a long time window can better capture the long-term characteristics of these noises, reduce the interference of instantaneous noise, and improve the stability of detection; a long time window can smooth instantaneous fluctuations and reduce false detections and missed detections caused by instantaneous energy changes. For example, in mechanical noise detection, a long time window can effectively filter out instantaneous noise when a machine tool starts and stops; for relatively stable noise, the use of a long time window can reduce the processing frequency, reduce the consumption of computing resources, and improve the overall efficiency of the system;

[0073] For noise signals of the instantaneous noise type, a short time window scale is used as the time scale window for noise detection; instantaneous noise changes rapidly, and the use of a short time window can quickly capture these noise changes and improve the real-time performance of detection. For example, in an environment with large changes in wind speed, a short time window can detect changes in wind noise in a timely manner to avoid delays; a short time window can capture the instantaneous characteristics of noise, which is especially important for burst noise. For example, knocking sounds or screams usually show high peak values ​​in a short period of time, and a short time window can effectively detect these instantaneous characteristics; the use of a short time window can speed up the response of the system, enabling the system to quickly adjust the gain control strategy and improve the clarity of the speech signal. For example, in a wind noise environment, a short time window can quickly reduce the gain of the speech signal to avoid interference from wind noise on the speech.

[0074] The second purpose of this embodiment is to provide a method for an industrial Internet software automation testing system, including the following method steps:

[0075] S1, obtaining a noise signal, and extracting the spectrum characteristics and time characteristics of the noise signal;

[0076] S2. Classify the noise signal according to the spectral characteristics and time characteristics of the noise signal to generate different types of noise signals; select different time window scales according to different types of noise signals to perform noise detection on the noise signal, wherein:

[0077] According to the frequency spectrum characteristics and time characteristics of the noise signal, the sizes of different time window scales are calculated, wherein the different time window scales include a long time window scale and a short time window scale.

[0078] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A remote voice control system for a numerical control system, characterized in that: The method comprises a feature extraction module (100), a noise classification and detection module (200) and a time window scale determination module (300), wherein: The feature extraction module (100) acquires a noise signal and extracts spectrum features and time features of the noise signal, wherein the spectrum features include spectrum amplitude and spectrum energy distribution, and the time features include average energy, energy variance and zero crossing rate; The noise classification and detection module (200) classifies the noise signal according to the spectral characteristics and time characteristics of the noise signal to generate different types of noise signals; according to different types of noise signals, different time window scales are selected to perform noise detection on the noise signal, wherein: The time window scale determination module (300) calculates the sizes of different time window scales according to the spectral characteristics and time characteristics of the noise signal, wherein the different time window scales include a long time window scale and a short time window scale; performs noise detection on the noise signal according to the different time window scales, so as to improve the accuracy of remote voice control; wherein the process of determining the size of the short time window scale specifically includes: According to the average energy in the temporal characteristics of the noise signal and zero crossing rate , calculate the size of the short time window scale, the calculation formula is as follows: ,in Indicates the size of the short time window; and Represents the average energy and zero crossing rate The weight coefficient is determined by the optimization algorithm; is a positive number to avoid the denominator being zero; The process of determining the size of the long time window specifically includes: According to the energy variance in the temporal characteristics of the noise signal The band energy distribution of the spectral characteristics is used to calculate the long-term window size. The calculation formula is as follows: ,in Indicates the size of the long time window; and Represents the energy variance and weight coefficients of frequency band energy distribution of spectral characteristics; A positive number to avoid The function's argument is zero; Indicates the low-frequency band energy given in the frequency band energy distribution; Indicates the high-frequency band energy in the frequency band energy distribution.

2. The remote voice control system of the numerical control system according to claim 1, characterized in that: The feature extraction module (100) uses a multi-microphone array to collect noise signals in the numerical control system environment.

3. The remote voice control system of the numerical control system according to claim 2, characterized in that: The feature extraction module (100) divides the collected noise signal into a plurality of fixed time windows, performs Fourier transform on the noise signal in each fixed time window, obtains the spectrum amplitude in the time window, and calculates the spectrum energy distribution of the spectrum amplitude in different frequency bands, the frequency bands including low frequency band, mid frequency band and high frequency band; and uses the spectrum amplitude and spectrum energy distribution as spectrum features.

4. The remote voice control system of the numerical control system according to claim 3, characterized in that: The process of extracting time features by the feature extraction module (100) is as follows: In each fixed time window, calculate the average energy of the noise signal , the formula is as follows: ,in is the number of sampling points in a fixed time window, It is The noise signal amplitude of each sampling point; In each fixed time window, calculate the energy variance of the noise signal , the formula is as follows: ； In each fixed time window, calculate the zero crossing rate of the noise signal , the formula is as follows: ,in is a sign function; The average energy , Energy variance and zero crossing rate As a time feature.

5. The remote voice control system of the numerical control system according to claim 4, characterized in that: The noise classification and detection module (200) comprises a noise classification unit (201), and the process of generating different types of noise signals by the noise classification unit (201) specifically comprises: The extracted spectral features and time features are combined into a feature vector as a fused feature vector; Different types of noise signal samples are collected from public databases and used as training data for the classification model. The sample types include stable noise and transient noise. According to the classification model structure, the classification model is trained using the training data to generate a trained noise signal classification model; The trained noise signal classification model is used to classify the fused feature vector to generate different types of noise signals.

6. The remote voice control system of the numerical control system according to claim 5, characterized in that: The noise classification unit (201) selects a convolutional neural network as a classification model for noise signals, and designs a structure of the classification model including an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer, wherein the input layer receives training data; each convolution layer contains a plurality of convolution kernels for extracting local features of the training data; the pooling layer is used to reduce feature dimensions and reduce the amount of calculation; the fully connected layer is used to map the extracted features to different noise categories; and the output layer outputs the category of the noise signal.

7. The remote voice control system of the numerical control system according to claim 6, characterized in that: The noise classification and detection module (200) comprises a noise detection unit (202), wherein the noise detection unit (202) selects different time window scales according to different types of noise signals to perform noise detection on the noise signal, specifically comprising: For the noise signal of the stable noise type, a long time window scale is used as the time scale window for noise detection; For noise signals of the instantaneous noise type, a short time window scale is used as the time scale window for noise detection.

8. A method for using a remote voice control system comprising a numerical control system according to any one of claims 1 to 7, characterized in that: The method comprises the following steps: S1. Obtain a noise signal and extract the spectral characteristics and time characteristics of the noise signal; S2. Classify the noise signal according to the spectral characteristics and time characteristics of the noise signal to generate different types of noise signals; select different time window scales according to different types of noise signals to perform noise detection on the noise signal, wherein: According to the frequency spectrum characteristics and time characteristics of the noise signal, the sizes of different time window scales are calculated, wherein the different time window scales include a long time window scale and a short time window scale.

Images