Speech recognition method and device for surgical power device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a voice recognition method and device for an operation power device, computer equipment, a computer readable storage medium and a computer program product, relates to the technical field of voice recognition, and can improve the control efficiency of the operation power device in the operation process. The method comprises the following steps: acquiring a control command voice signal of a surgical power device, and decomposing and modulating the control command voice signal to obtain a plurality of voice signal components corresponding to frequency modes in different frequency ranges; carrying out instantaneous feature extraction on each component, and carrying out feature normalization processing to generate a corresponding current sub-signal; iteratively adjusting each sub-signal by taking the minimum bandwidth as a target, and determining a target sub-signal and a center frequency thereof according to an iteration result; according to the center frequency, periodic features of the target sub-signals are extracted, and feature signals are generated; and inputting the characteristic signals into a trained voice classification model to finally obtain a recognition result for executing a corresponding control command.

Description

Technical Field

[0001] The present application relates to the field of speech recognition technology, and in particular to a speech recognition method, apparatus, computer equipment, computer-readable storage medium, and computer program product for a surgical power device. Background Art

[0002] The operation of surgical power devices is often involved in the surgical process. Traditional operations mostly rely on the operator to adjust the operation of the device through manual buttons, knobs or pedals. The advantages of this control method are that it is relatively simple and reliable. After years of clinical practice, it has been widely used.

[0003] However, during the operation, the operator is highly focused on the operation and often has difficulty in quickly completing the manual control of the equipment. Especially in an emergency, the operator may not be able to adjust the equipment in time, resulting in related risks. It can be seen that the control efficiency of the surgical power device in the related technology is low. Summary of the invention

[0004] Based on this, it is necessary to provide a speech recognition method, device, computer equipment, computer-readable storage medium and computer program product for a surgical power device that can improve the control efficiency of the surgical power device in order to address the above technical problems.

[0005] In a first aspect, the present application provides a speech recognition method for a surgical power device, comprising:

[0006] Acquire a control command voice signal input for the surgical power device;

[0007] Performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges;

[0008] Performing instantaneous feature extraction on each of the speech signal components, and normalizing the instantaneous feature extraction results of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components;

[0009] Taking minimizing the bandwidth of each current sub-signal as the optimization goal, iteratively adjusting each current sub-signal, and determining each target sub-signal and the center frequency corresponding to each target sub-signal according to the iteration result;

[0010] According to the center frequency of each of the target sub-signals, periodic feature extraction is performed on each of the target sub-signals to obtain a feature signal corresponding to each of the target sub-signals;

[0011] Each of the feature signals is input into a trained speech classification network to obtain a control command recognition result output by the speech classification network.

[0012] In one embodiment, the instantaneous feature extraction of each of the speech signal components and normalization of the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components includes:

[0013] Calculating the time change rate of each of the speech signal components to extract the instantaneous features corresponding to the speech signal components;

[0014] The amplitude values ​​of the instantaneous features are adjusted to the same preset range to obtain current sub-signals corresponding to the speech signal components.

[0015] In one embodiment, minimizing the bandwidth of each current sub-signal as the optimization goal, iteratively adjusting each current sub-signal, and determining each target sub-signal and the center frequency corresponding to each target sub-signal according to the iteration result include:

[0016] Calculating the frequency distribution of the current sub-signal, and evaluating the bandwidth of the current sub-signal according to the frequency distribution to obtain an evaluation result; the evaluation result indicates whether the bandwidth meets the optimization goal and the frequency distribution adjustment direction;

[0017] According to the evaluation result, adjusting the frequency distribution of the current sub-signal and the center frequency of the current sub-signal to obtain an updated current sub-signal;

[0018] Return to the step of calculating the frequency distribution of the current sub-signal until the evaluation result of the current sub-signal meets the optimization target, determine the current sub-signal as the target sub-signal, and obtain the center frequency corresponding to the target sub-signal.

[0019] In one embodiment, performing period feature extraction on each target sub-signal according to the center frequency of each target sub-signal to obtain a feature signal corresponding to each target sub-signal includes:

[0020] orthogonally splitting the target sub-signal corresponding to the center frequency to obtain a first split signal and a second split signal, and extracting periodic features from the first split signal and the second split signal to obtain a plurality of first periodic features of the first split signal and a plurality of second periodic features of the second split signal;

[0021] Combining and analyzing the plurality of first periodic features to obtain a periodic characteristic signal of a first oscillation mode, and combining and analyzing the plurality of second periodic features to obtain a periodic characteristic signal of a second oscillation mode;

[0022] The characteristic signal corresponding to each of the target sub-signals is obtained according to the periodic characteristic signal of the first oscillation mode and the periodic characteristic signal of the second oscillation mode corresponding to each of the target sub-signals.

[0023] In one embodiment, the speech classification model is trained by the following steps, including:

[0024] Acquire a historical feature signal corresponding to a sample of a control command voice signal input to a surgical power device, and fuse the historical feature signal into a sample two-dimensional time-frequency graph; the sample two-dimensional time-frequency graph includes a period feature and a frequency feature of the historical feature signal;

[0025] Inputting the sample two-dimensional time-frequency graph into a speech classification model to be trained, extracting local features and overall features of the sample two-dimensional time-frequency graph by the speech classification model, and outputting a command recognition result for the control command speech signal sample according to the local features and the overall features;

[0026] According to the command recognition result of the control command voice signal sample, the model parameters of the voice classification model are adjusted, and when the training end condition is met, the trained voice classification model is obtained.

[0027] In one embodiment, after inputting each of the feature signals into a trained speech classification model to obtain a control command recognition result output by the speech classification model, the method further includes:

[0028] Performing semantic understanding on the control command voice signal and converting the control command voice signal into a control command text;

[0029] Matching the control command text with a plurality of command corpora in a pre-established command corpus to obtain a matching degree between the control command text and each of the plurality of command corpora;

[0030] According to each of the matching degrees, the command corpus in the command corpus is updated.

[0031] In one embodiment, the performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges includes:

[0032] Decomposing the control command voice signal into a plurality of voice modal signals, and extracting timing information of each of the voice modal signals;

[0033] Determining the center frequency of each of the sub-signals according to the timing information;

[0034] According to the center frequency, a frequency shift operation is performed on the sub-signal to obtain the speech signal components under multiple frequency modes corresponding to different center frequency ranges.

[0035] In a second aspect, the present application also provides a speech recognition device for a surgical power device, comprising:

[0036] A voice signal acquisition module, used to obtain a control command voice signal input to the surgical power device;

[0037] A signal decomposition and modulation module, used for performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges;

[0038] An instantaneous feature extraction and normalization module extracts instantaneous features from each of the speech signal components, and normalizes the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components;

[0039] A bandwidth optimization module, configured to iteratively adjust each of the current sub-signals with minimizing the bandwidth of each of the current sub-signals as the optimization target, and determine each of the target sub-signals and the center frequency corresponding to each of the target sub-signals according to the iteration result;

[0040] A periodic feature extraction module, used to extract periodic features of each target sub-signal according to the center frequency of each target sub-signal, so as to obtain a feature signal corresponding to each target sub-signal;

[0041] The speech classification and recognition module is used to input each of the feature signals into a trained speech classification model to obtain a control command recognition result output by the speech classification model.

[0042] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0043] Acquire a control command voice signal input for the surgical power device;

[0044] Performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges;

[0045] Extracting instantaneous features of each of the speech signal components, and normalizing the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components;

[0046] Taking minimizing the bandwidth of each current sub-signal as the optimization goal, iteratively adjusting each current sub-signal, and determining each target sub-signal and the center frequency corresponding to each target sub-signal according to the iteration result;

[0047] According to the center frequency of each of the target sub-signals, periodic feature extraction is performed on each of the target sub-signals to obtain a feature signal corresponding to each of the target sub-signals;

[0048] Each of the feature signals is input into a trained speech classification model to obtain a control command recognition result output by the speech classification model.

[0049] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0050] Acquire a control command voice signal input for the surgical power device;

[0051] Performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges;

[0052] Extracting instantaneous features of each of the speech signal components, and normalizing the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components;

[0053] Taking minimizing the bandwidth of each current sub-signal as the optimization goal, iteratively adjusting each current sub-signal, and determining each target sub-signal and the center frequency corresponding to each target sub-signal according to the iteration result;

[0054] According to the center frequency of each of the target sub-signals, periodic feature extraction is performed on each of the target sub-signals to obtain a feature signal corresponding to each of the target sub-signals;

[0055] Each of the feature signals is input into a trained speech classification model to obtain a control command recognition result output by the speech classification model.

[0056] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0057] Acquire a control command voice signal input for the surgical power device;

[0058] Performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges;

[0059] Extracting instantaneous features of each of the speech signal components, and normalizing the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components;

[0060] Taking minimizing the bandwidth of each current sub-signal as the optimization goal, iteratively adjusting each current sub-signal, and determining each target sub-signal and the center frequency corresponding to each target sub-signal according to the iteration result;

[0061] According to the center frequency of each of the target sub-signals, periodic feature extraction is performed on each of the target sub-signals to obtain a feature signal corresponding to each of the target sub-signals;

[0062] Each of the feature signals is input into a trained speech classification model to obtain a control command recognition result output by the speech classification model.

[0063] In the above-mentioned speech recognition method for surgical power devices, a control command speech signal input to the surgical power device is obtained; the control command speech signal is decomposed and modulated to obtain speech signal components corresponding to multiple frequency modes in different frequency ranges, and speech signal components corresponding to the control command speech signal in frequency modes in different frequency ranges are obtained; instantaneous features are extracted for each speech signal component, and the instantaneous features of each extracted speech signal component are normalized to obtain the current sub-signal corresponding to each speech signal component; with minimizing the bandwidth of each current sub-signal as the optimization goal, each current sub-signal is iteratively adjusted, and each target sub-signal and the center frequency corresponding to each target sub-signal are determined according to the iteration result; according to the center frequency of each target sub-signal, periodic features are extracted for each target sub-signal to obtain feature signals corresponding to each of the target sub-signals; each feature signal is input into a trained speech classification model to obtain a control command recognition result output by the speech classification model. In this application, accurate analysis of different frequency modes is achieved by decomposing the modulated voice signal and extracting instantaneous features; subsequently, through iterative optimization of bandwidth minimization and periodic feature extraction, the system effectively reduces noise interference and frequency redundancy, and the characteristic signal finally generated ensures that the trained voice classification model can accurately identify control commands, thereby improving the voice control accuracy and response speed of the surgical power device, ensuring high efficiency during the operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0065] FIG1 is a schematic flow chart of a method for speech recognition of a surgical power device in one embodiment;

[0066] FIG2 is a flow chart of a voice recognition method for a surgical power device in another embodiment;

[0067] FIG3 is a structural block diagram of a speech recognition device for a surgical power device in one embodiment;

[0068] FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0070] In one embodiment, as shown in FIG1 , a speech recognition method for a surgical power device is provided. This embodiment is illustrated by applying the method to a terminal. In some examples, the terminal is a surgical power device. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0071] Step S101, obtaining a control command voice signal input to a surgical power device.

[0072] The control command voice signal can be understood as a voice instruction inputted by voice, which is intended to control the surgical power device to perform a specific operation. These instructions may include but are not limited to one or more of the following: change the speed, adjust the direction, stop the operation. The signal is transmitted to the computer terminal through the audio acquisition device.

[0073] Optionally, when the terminal obtains the control command voice signal, the real-time audio data stream is saved in the cache, and the obtained audio signal is preprocessed, including steps such as denoising and filtering.

[0074] Optionally, the terminal can integrate multiple audio input devices, such as an array with multiple microphones, so as to improve the collection effect of voice signals through multi-channel input. In this case, the terminal selects the audio input with the highest signal quality as the control command voice signal and ignores other input channels. This can effectively reduce noise interference in the signal and further improve the accuracy of the voice signal.

[0075] Step S102: performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes within different frequency ranges.

[0076] Among them, frequency mode refers to the performance of speech signals in different frequency ranges. Different frequency modes represent different components of speech signals. For example, the low-frequency part contains the basic frequency of speech, while the high-frequency part contains more subtle speech features. By analyzing these modes, various characteristics of the signal can be obtained.

[0077] Optionally, the terminal decomposes the preprocessed voice signal based on a preset number of modal decompositions, and on the basis of the decomposed signal, the terminal further modulates each frequency mode. The purpose of modulation is to combine the time domain characteristics and frequency domain characteristics of the voice signal. Through modulation, the terminal can further extract the frequency components and corresponding time changes in the voice signal, so that the system can more comprehensively grasp the changes in the voice signal when processing the voice.

[0078] Through decomposition and modulation, the terminal obtains voice signal components under multiple frequency modes. These signal components contain the performance information of the voice signal in different frequency ranges. The terminal stores these voice signal components as digital data and prepares for subsequent feature extraction and analysis. Each signal component represents a specific frequency range, and its content includes the energy, amplitude and other important features of the voice signal within the frequency range.

[0079] Optionally, when higher resolution is required, the terminal can increase the number of frequency modes to be decomposed so as to capture more details in the voice signal. In a noisy environment, the terminal can adjust the modulation algorithm to enhance the noise resistance of the signal component.

[0080] Step S103: extract instantaneous features of each speech signal component, and normalize the extracted instantaneous features of each speech signal component to obtain a current sub-signal corresponding to each speech signal component.

[0081] Among them, instantaneous feature extraction is to extract the instantaneous properties of the signal from the signal components, which can be instantaneous amplitude, instantaneous frequency, instantaneous phase, etc. These features can reflect the details of the signal that changes rapidly in time, which helps the system to identify the state changes of the signal at a certain moment when processing the voice signal, thereby improving the system's sensitivity to voice commands.

[0082] Optionally, the terminal may perform multi-level instantaneous feature extraction for different speech signal components to obtain instantaneous information of more dimensions. In a specific embodiment, the terminal may increase the extraction of high-order instantaneous features as needed to further improve the recognition accuracy.

[0083] Optionally, when processing different types of control command voice signals, the terminal can dynamically adjust the normalization strategy according to the specific characteristics of the signal. In a specific embodiment, for low-frequency signals, a normalization algorithm with excellent averaging performance can be selected to ensure that the low-frequency signal component is more stable after normalization; for high-frequency signals, a normalization algorithm with excellent speed performance can be selected to shorten the normalization processing time of high-frequency signals with fewer signal features.

[0084] Step S104, taking minimizing the bandwidth of each current sub-signal as the optimization goal, iteratively adjusting each current sub-signal, and determining each target sub-signal and the center frequency corresponding to each target sub-signal according to the iteration result.

[0085] Bandwidth is the width of the signal spectrum in the frequency domain, indicating the range of the frequency component distribution of the signal. The optimization goal of bandwidth minimization is to minimize the spectrum range occupied by the current sub-signal while keeping the signal characteristics unchanged. By minimizing the bandwidth, the signal concentration and the accuracy of frequency analysis can be improved, and noise interference can be reduced.

[0086] The center frequency refers to the concentrated position of the frequency components in the target sub-signal, representing the main frequency characteristics of the signal component in the frequency range. Each target sub-signal has a corresponding center frequency.

[0087] Optionally, the current sub-signal has a wide spectrum range, resulting in some frequency components being redundant or noise interference. The terminal first calculates the initial bandwidth of each current sub-signal and determines its distribution range in the frequency domain. The bandwidth is calculated based on the spectrum density of each sub-signal. By analyzing the extension degree of the sub-signal in the frequency domain, the terminal can derive the bandwidth size of the current sub-signal. The wider the bandwidth, the more frequency components the signal contains, but this also leads to an increase in irrelevant information in the signal. Therefore, the goal of bandwidth optimization is to concentrate the signal into a narrower frequency range, thereby improving the effectiveness of the signal.

[0088] After multiple iterations of adjustment, the bandwidth of the current sub-signal gradually decreases, and the system finally obtains the optimal bandwidth of each signal component. At this point, the current sub-signal is converted into a target sub-signal. The target sub-signal has a narrower bandwidth, a more concentrated frequency distribution, and contains the core information in the voice signal. After the target sub-signal is determined, the terminal will further analyze the frequency distribution of each sub-signal and extract the center frequency of the signal.

[0089] Optionally, the terminal may dynamically adjust the bandwidth optimization standard according to the characteristics of the signal. In a specific embodiment, for a high-noise environment, the terminal may set a more stringent bandwidth minimization standard to extract a more accurate speech signal under high-noise conditions.

[0090] Step S105 , extracting periodic features of each target sub-signal according to the center frequency of each target sub-signal to obtain a feature signal corresponding to each target sub-signal.

[0091] Optionally, the terminal first determines the target sub-signal corresponding to the center frequency based on the center frequency of the target sub-signal, and performs periodic detection on the signal. The purpose of periodic detection is to determine the repetitive pattern of the signal in time, such as the rise or fall of the tone, the repetition of syllables, etc. By analyzing the periodicity of the signal, the terminal can identify the repetitive pattern in the speech signal and further extract relevant features. After the periodic feature extraction is completed, the terminal combines the extracted periodic features to generate the final feature signal.

[0092] Optionally, the terminal may perform multi-scale periodic feature extraction on the target sub-signal, thereby capturing long-term and short-term periodic changes in the speech signal by analyzing the periodic features of the signal at different time scales.

[0093] Step S106, input each feature signal into the trained speech classification model to obtain the control command recognition result output by the speech classification model.

[0094] Optionally, the terminal uses the feature signal as input and transmits it to a pre-trained speech classification model. The input format of the feature signal matches the input requirements of the classification model and can be in the form of a multi-dimensional vector or matrix, representing the performance of the speech signal in different feature dimensions. Depending on the application scenario, the architecture of the speech classification model can be flexibly adjusted.

[0095] Optionally, the terminal can adopt a multi-model fusion approach, that is, by combining the classification results of multiple models, a more accurate control command recognition result can be obtained. Each model focuses on different feature dimensions or signal types, and ultimately outputs a comprehensive classification decision through a voting mechanism or weighted average.

[0096] In the above-mentioned speech recognition method for surgical power devices, a control command speech signal input to the surgical power device is obtained; the control command speech signal is decomposed and modulated to obtain speech signal components corresponding to the control command speech signal in frequency modes in different frequency ranges; instantaneous features are extracted for each speech signal component, and the instantaneous features of each extracted speech signal component are normalized to obtain a current sub-signal corresponding to each speech signal component; each current sub-signal is iteratively adjusted with minimizing the bandwidth of each current sub-signal as the optimization goal, and each target sub-signal and the center frequency corresponding to each target sub-signal are determined according to the iteration result; based on the center frequency of each target sub-signal, periodic features are extracted for each target sub-signal to obtain a feature signal corresponding to each target sub-signal; each feature signal is input into a trained speech classification model to obtain a control command recognition result output by the speech classification model. In this embodiment, by decomposing the modulated voice signal and extracting instantaneous features, accurate analysis of different frequency modes is achieved; then, through iterative optimization of bandwidth minimization and periodic feature extraction, the system effectively reduces noise interference and frequency redundancy. The characteristic signal finally generated ensures that the trained voice classification model can accurately identify control commands, thereby improving the voice control accuracy and response speed of the surgical power device, ensuring high efficiency during the operation.

[0097] In an exemplary embodiment, performing signal decomposition and modulation on a control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges includes:

[0098] The control command voice signal is decomposed into multiple voice mode signals, and the timing information of each voice mode signal is extracted; the center frequency of each sub-signal is determined according to the timing information; and the sub-signal is subjected to frequency shift operation according to the center frequency to obtain voice signal components under multiple frequency modes corresponding to different center frequency ranges.

[0099] Specifically, the terminal uses a decomposition algorithm to Split into multiple sub-signals in different frequency ranges, namely speech mode signals , its mathematical expression is formula (1):

[0100]

[0101] The terminal extracts the time series features in the signal through the Hilbert change, which can be expressed by formula (2). The terminal can analyze the changes of the speech signal in a short time window, which is very helpful for capturing the stress, pauses and dynamic changes of the speech signal. These time series features will provide a basis for the subsequent frequency feature analysis.

[0102]

[0103] The terminal uses fast Fourier transform (FFT) to perform frequency domain conversion on each voice modal signal. Through FFT, the time domain signal is converted into a frequency domain signal, and the terminal can obtain the frequency distribution of each modal signal in the frequency domain. The FFT result can show the energy intensity of each frequency component. By analyzing the FFT result, the terminal can accurately determine the center frequency of each modal signal.

[0104] After the center frequency is determined, the terminal performs center modulation on the signal. The center modulation is performed by Convolution is completed, the formula represents the center frequency, The time terminal adds the frequency characteristics of the center frequency to the signal, so that the signal shows a more concentrated frequency distribution in the frequency domain, thereby obtaining the voice signal components corresponding to the control command voice signal in the frequency mode of different center frequency ranges.

[0105] In this embodiment, the key information in the signal is accurately extracted and processed through steps such as timing feature extraction, fast Fourier transform and center modulation. This process not only enhances the ability to capture the timing and frequency features of the signal, but also improves the analysis effect of the signal by optimizing the bandwidth and frequency distribution, so that the voice command can be more accurately recognized and executed by the system.

[0106] In an exemplary embodiment, instantaneous features are extracted from each speech signal component, and the extracted instantaneous features of each speech signal component are normalized to obtain a current sub-signal corresponding to each speech signal component, including:

[0107] The time change rate of each speech signal component is calculated to extract the instantaneous features corresponding to the speech signal components; the amplitude values ​​of the instantaneous features are adjusted to the same preset range to obtain the current sub-signals corresponding to each speech signal component.

[0108] Specifically, the terminal performs instantaneous feature and normalization processing on the obtained speech component signal, which can be achieved by formula (3) in some examples.

[0109]

[0110] After receiving the voice signal component, the terminal can obtain the time change rate of each component, that is, the instantaneous change rate, by performing derivative operations on the timing information of the voice signal component. The instantaneous change rate reflects the change of the signal in the time dimension and captures the short-term characteristics of the voice signal, such as changes in amplitude and frequency. For the recognition of control command voice, the instantaneous amplitude can be used to distinguish the tone intensity and stress of different voice commands. The instantaneous frequency reflects the change speed of a certain frequency component in the control command voice, and can be used to capture high-frequency or low-frequency changes in the voice signal. This is particularly important for distinguishing different syllables, words or voice commands in the voice signal.

[0111] After the instantaneous feature extraction is completed, the terminal normalizes the extracted instantaneous amplitude, instantaneous frequency and other information. The purpose of normalization is to standardize the instantaneous feature values ​​of different speech signal components so that they are in the same numerical range. This step eliminates the deviation caused by the difference in amplitude or frequency of different signal components, making the subsequent feature analysis process more stable and consistent.

[0112] After normalization, the terminal generates a current sub-signal corresponding to each speech signal component.

[0113] In this embodiment, by calculating the time rate of change of the speech signal components, the instantaneous characteristics of the signal can be accurately captured, and the precise analysis of the dynamic changes of the speech signal can be achieved. Subsequently, the amplitude of the instantaneous characteristics is normalized to ensure that the characteristic values ​​of different signal components are processed within the same range, reducing the analysis error caused by the difference in characteristic scales. Ultimately, this process achieves the purpose of generating a standardized current sub-signal, providing high-quality input for subsequent signal processing and classification, and improving the accuracy and consistency of voice command recognition.

[0114] In an exemplary embodiment, minimizing the bandwidth of each current sub-signal is taken as an optimization goal, each current sub-signal is iteratively adjusted, and each target sub-signal and a center frequency corresponding to each target sub-signal are determined according to the iteration result, including:

[0115] Calculate the frequency distribution of the current sub-signal, evaluate the bandwidth of the current sub-signal according to the frequency distribution, and obtain an evaluation result; the evaluation result indicates whether the bandwidth meets the optimization goal and the frequency distribution adjustment direction; according to the evaluation result, adjust the frequency distribution of the current sub-signal and the center frequency of the current sub-signal to obtain an updated current sub-signal; return to execute the step of calculating the frequency distribution of the current sub-signal until the evaluation result of the current sub-signal meets the optimization goal, determine the current sub-signal as the target sub-signal, and obtain the center frequency corresponding to the target sub-signal.

[0116] Specifically, after receiving the current sub-signal, the terminal performs frequency domain analysis on the signal to calculate its frequency distribution. The frequency distribution reflects the energy distribution of each frequency component in the signal. The terminal can use frequency domain analysis methods such as FFT to obtain the performance of the current sub-signal in the frequency domain.

[0117] The terminal evaluates the rationality of the bandwidth by calculating the distribution width of the frequency components in the current sub-signal. For example, the terminal can calculate the coverage of the frequency component energy in the signal through the L2 paradigm, which can be expressed by formula (4) to determine whether the bandwidth meets the optimization goal of the system.

[0118]

[0119] After the evaluation is completed, the terminal compares the bandwidth evaluation result with the predetermined optimization target. If the bandwidth meets the optimization target (for example, the bandwidth is narrow and the frequency components are concentrated), the signal does not need to be further adjusted; if the bandwidth is too wide (meaning that the frequency distribution of the signal is more dispersed), the frequency distribution of the current sub-signal needs to be adjusted.

[0120] While adjusting the frequency distribution, the terminal will also optimize the center frequency of the current sub-signal. The terminal will calculate the spectrum peak position of the current sub-signal and fine-tune the center frequency based on the bandwidth evaluation result to more accurately reflect the frequency characteristics of the signal.

[0121] After the adjustment is completed, the terminal will calculate the frequency distribution of the current sub-signal again to check the effect of the center frequency adjustment and bandwidth optimization. If the bandwidth still does not meet the optimization target, the terminal will continue to iterate until the bandwidth meets the predetermined target and the frequency distribution and center frequency of the signal are stable.

[0122] When the bandwidth optimization meets the requirements, the terminal determines the current sub-signal as the target sub-signal and records the center frequency of the target sub-signal as the basis for subsequent processing and feature extraction.

[0123] In this embodiment, the frequency dispersion and low-frequency noise interference problems are effectively solved through bandwidth optimization and dynamic frequency adjustment. The optimization of frequency distribution makes the frequency components of the signal more concentrated, reduces redundant frequency information, and especially ensures that the main frequency characteristics are retained by reducing low-frequency noise. This optimization process improves the signal processing accuracy of the system and the accuracy of voice command recognition, and enhances the robustness of the system in complex environments.

[0124] In an exemplary embodiment, periodic feature extraction is performed on each target sub-signal according to the center frequency of each target sub-signal to obtain a feature signal corresponding to each target sub-signal, including:

[0125] The target sub-signal corresponding to the center frequency is orthogonally split to obtain a first split signal and a second split signal, and periodic features are extracted from the first split signal and the second split signal to obtain multiple first periodic features of the first split signal and multiple second periodic features of the second split signal; the multiple first periodic features are combined and analyzed to obtain a periodic characteristic signal of the first oscillation mode, and the multiple second periodic features are combined and analyzed to obtain a periodic characteristic signal of the second oscillation mode; the characteristic signal corresponding to each target sub-signal is obtained according to the periodic characteristic signal of the first oscillation mode and the periodic characteristic signal of the second oscillation mode corresponding to each target sub-signal.

[0126] Specifically, the terminal extracts the periodic features according to formula (5). The term represents the first split signal of the target sub-signal. Specifically, the corresponding phase and amplitude periodic characteristics are extracted from the cosine component of the target sub-signal, and then the first oscillation mode under the current cosine component is determined. The periodic characteristics under the first oscillation mode are integrated and weighted to obtain the periodic characteristics of the first split signal.

[0127] In the formula The term represents the second split signal of the target sub-signal. This section describes the periodic feature extraction of the signal through the sinusoidal component. Items are similar, The partial integration operation represents a periodic analysis of the amplitude and phase of the signal. Specifically, the corresponding phase and amplitude periodic characteristics are extracted from the cosine angle of the target sub-signal, and then the second oscillation mode under the current sinusoidal component is determined. The periodic characteristics under the second oscillation mode are integrated and weighted to obtain the periodic characteristics of the second split signal.

[0128]

[0129] In this embodiment, the target sub-signal is decomposed into a first split signal and a second split signal, and the amplitude and phase period characteristics of the signal are extracted respectively. By integrating and analyzing the period characteristics of these two parts of the signal, the terminal generates a first oscillation mode and a second oscillation mode, and finally combines the two to obtain the characteristic signal of the target sub-signal. This process effectively solves the problem of capturing complex amplitude and phase changes in speech signals, ensures the accurate extraction of signal period characteristics, and provides high-precision feature input for subsequent speech recognition.

[0130] In an exemplary embodiment, the speech classification model is trained by the following steps, including:

[0131] Obtain historical feature signals corresponding to control command voice signal samples input to the surgical power device, and fuse the historical feature signals into a sample two-dimensional time-frequency diagram; the sample two-dimensional time-frequency diagram includes periodic features and frequency features of the historical feature signals; input the sample two-dimensional time-frequency diagram into a speech classification model to be trained, extract local features and overall features of the sample two-dimensional time-frequency diagram by the speech classification model, and output command recognition results for the control command voice signal samples based on the local features and overall features; adjust model parameters of the speech classification model based on the command recognition results of the control command voice signal samples, and obtain the trained speech classification model when the training end conditions are met.

[0132] Specifically, the terminal extracts historical feature signals from previously collected surgical power device control command voice signal samples. These historical feature signals include the periodic and frequency features of the signal, which are information accumulated during multiple voice signal processing processes in the past. The terminal integrates these historical feature signals into a two-dimensional time-frequency graph of the sample.

[0133] The two-dimensional time-frequency graph is transmitted as input to the speech classification model to be trained. The model extracts the local and overall features of the sample by analyzing the information in the time-frequency graph. The local features can reflect the changes of the signal in a short period of time, such as short-term periodic fluctuations, frequency changes, etc. The overall features can capture the global trend of the signal, such as overall periodic fluctuations or frequency distribution.

[0134] Through comprehensive analysis of local and overall features, the speech classification model outputs command recognition results for control command speech signal samples. These results are specific operation commands for surgical equipment, such as "start", "stop" or "adjust speed". Based on the recognition results, the terminal dynamically adjusts the model parameters to ensure that the model can continuously optimize its weight distribution as the recognition accuracy improves. When the recognition accuracy of the model reaches the preset standard, the training process ends and a trained speech classification model is finally obtained.

[0135] Exemplarily, the speech classification model selects a lightweight global-local feature fusion ViT (Vision Transformer) model. On the basis of obtaining a two-dimensional time-frequency graph of the feature signal, each image is divided into image blocks of fixed size, and each image block is flattened into a vector. After each image block is flattened, it is mapped to an embedding space of fixed dimension through a linear transformation. Afterwards, in order to maintain the relative position information of the image block in space, the ViT model also adds a position code to each image block. The image block embedding with position code is sent to the Transformer encoder for processing. The Transformer encoder consists of a multi-layer self-attention mechanism and a feedforward neural network. Here, the self-attention mechanism calculates the correlation between each image block and other image blocks, and can capture the global features in the image while capturing local features. The output of the Transformer encoder is used for image classification tasks through a classification head. The ViT model adds an additional classification tag to the image block embedding and uses the output of the tag as the representation of the final classification. This classification tag aggregates a global feature from the features of the entire image for the overall classification of the image. During the training process, the back-propagation algorithm can be used to update the parameters of the model through the optimizer. When the recognition accuracy of the model reaches the preset standard, the training process ends, and finally a trained lightweight global-local feature fusion ViT (Vision Transformer) model is obtained.

[0136] In this embodiment, by obtaining the historical characteristic signals of the control command voice signal samples input to the surgical power device and generating a sample two-dimensional time-frequency diagram containing periodic characteristics and frequency characteristics, a comprehensive representation of the time-frequency characteristics of the voice signal is achieved, so that the voice classification model can capture both the local and overall characteristics of the signal, thereby enabling the voice classification model to have a stronger signal feature extraction capability. By adjusting the parameters of the model based on the recognition results, dynamic optimization is achieved during the model training process, ensuring that the model can continuously improve the recognition accuracy of complex control commands. Ultimately, the goal of improving the recognition accuracy and robustness of the voice classification model for the voice control commands of the surgical power device is achieved, so that the model can accurately and quickly respond to different voice commands and adapt to the efficient operation requirements in the surgical environment.

[0137] In an exemplary embodiment, after inputting each feature signal into a trained speech classification model to obtain a control command recognition result output by the speech classification model, the method further includes:

[0138] The control command voice signal is semantically understood and converted into a control command text; the control command text is matched with a plurality of command corpora in a pre-established command corpus to obtain a matching degree between the control command text and each of the plurality of command corpora; and the command corpora in the command corpus are updated according to each matching degree.

[0139] Specifically, after the speech classification model outputs the control command recognition result, the terminal performs semantic understanding on the speech signal of the control command. Semantic understanding can be to parse the meaning of the speech signal through the fast-whisper model and convert the speech signal of the control command into a structured text representation. The generated control command text is then matched with multiple command corpora in a pre-established command corpus. The command corpus stores standardized expressions of various control commands, and each corpus represents a specific control command.

[0140] The terminal calculates the matching degree between the control command text and each corpus in the command corpus through a semantic matching algorithm. Based on the matching result, the terminal updates the command corpus. The updating process can be based on a matching threshold. If the matching degree is lower than a preset value, the terminal considers the command to be a new command and adds it to the corpus. For existing commands, if they have a high matching degree with the new command text but there are some differences, the terminal will modify or expand them to accommodate more diverse inputs.

[0141] In this embodiment, by converting the control command voice signal into text and matching it with the corpus, the ability to accurately parse the voice command and dynamically update the corpus is achieved. This not only improves the accuracy of voice command recognition, but also enhances the system's adaptability to new commands or variant commands. Ultimately, through the continuous optimization of the corpus, the goal of improving the robustness and scalability of the voice recognition system in complex operating environments is achieved.

[0142] In order to enable those skilled in the art to better understand the above steps, the embodiment of the present application is illustrated below by using an example, but it should be understood that the embodiment of the present application is not limited to this.

[0143] As shown in Figure 2, the system acquires voice signals in the surgical environment. These signals are collected through terminal devices and include control commands input by doctors or surgical personnel, such as "start the device" and "adjust the speed". The preprocessing steps include denoising, predefinition, voice signal segmentation and endpoint detection to ensure the high quality of the input signal and remove background noise and irrelevant information.

[0144] The preprocessed speech signal is input into the variational mode decomposition (VMD) algorithm. The role of VMD is to decompose the signal into multiple intrinsic mode functions. The intrinsic mode function represents the characteristics of the speech signal in different frequency ranges. After that, the analysis signal of each intrinsic mode function is calculated, and these modes are mixed with exponential functions to transfer their frequencies to the baseband. Finally, the bandwidth of each intrinsic mode function is estimated by the squared L2 norm of the gradient. The optimization goal is to ensure that the frequency components of each mode are more concentrated and eliminate unnecessary frequency extension.

[0145] After VMD decomposition is completed, the decomposed signal is Hilbert transformed. By performing a periodic Hilbert transform on each intrinsic mode function signal output by VMD, the signal is analyzed using sine and cosine components, and the system generates an analytical signal and extracts the corresponding periodic features.

[0146] The signal processed by VMD and periodic Hilbert transform is converted into a two-dimensional time-frequency graph and provided as input to a lightweight global-local feature fusion ViT classification network to output the voice control command recognition result.

[0147] The system establishes a corpus of common control commands for surgical power devices. The corpus contains different voice commands and their corresponding text instructions, such as steering (forward, reverse, reciprocating), speed (increase, decrease), frequency (increase, decrease, how many hertz), flow (increase, decrease), etc. Based on the open source fast-whisper model, the system converts the recognized voice signal into text format. The fast-whisper model can provide object representation format (json, JavaScriptObject Notation), srt (SubRip Subtitle) subtitles with timestamp format or plain text format, which is convenient for further analysis of voice commands. The system matches and analyzes the control voice command with multiple standard commands in the corpus, and determines the control command to be executed based on the degree of match.

[0148] In this embodiment, the combined VMD and Hilbert techniques enable the system to accurately capture the time-frequency characteristics of voice signals in complex surgical environments, ensuring accurate recognition of voice commands. Compared with the traditional DNN-based i-vector recognition method, the recognition accuracy of the present invention on text-independent data sets is improved by 60%. In addition, combined with the use of the ViT model and corpus dictionary, the system not only improves the recognition efficiency of voice commands, but also adapts to different command variants, effectively reducing the labor cost in clinical operations and improving surgical efficiency.

[0149] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0150] Based on the same inventive concept, the embodiment of the present application also provides a speech recognition device for a surgical power device for implementing the speech recognition method for a surgical power device. The implementation solution provided by the device is similar to the implementation solution recorded in the above method, so the specific limitations in one or more speech recognition device embodiments for a surgical power device provided below can refer to the limitations of the speech recognition method for a surgical power device above, and will not be repeated here.

[0151] In an exemplary embodiment, as shown in FIG3 , a speech recognition device for a surgical power device is provided, comprising: a speech signal acquisition module 310, a signal decomposition and modulation module 320, an instantaneous feature extraction and normalization module 330, a bandwidth optimization module 340, a periodic feature extraction module 350, and a speech classification and recognition module 360, wherein:

[0152] The voice signal acquisition module 310 is used to obtain the control command voice signal input to the surgical power device;

[0153] The signal decomposition and modulation module 320 is used to perform signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges;

[0154] The instantaneous feature extraction and normalization module 330 performs instantaneous feature extraction on each of the speech signal components, and performs normalization processing on the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components;

[0155] The bandwidth optimization module 340 is used to iteratively adjust each of the current sub-signals with minimizing the bandwidth of each of the current sub-signals as the optimization target, and determine each of the target sub-signals and the center frequency corresponding to each of the target sub-signals according to the iteration result;

[0156] A periodic feature extraction module 350 is used to extract periodic features of each target sub-signal according to the center frequency of each target sub-signal to obtain a feature signal corresponding to each target sub-signal;

[0157] The speech classification and recognition module 360 ​​is used to input each of the feature signals into a trained speech classification model to obtain a control command recognition result output by the speech classification model.

[0158] In one embodiment, the signal decomposition and modulation module 320 is used to: decompose the control command voice signal into multiple voice mode signals and extract the timing information of each voice mode signal; determine the center frequency of each sub-signal based on the timing information; and perform frequency shift operations on the sub-signals based on the center frequency to obtain the voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges.

[0159] In one embodiment, the instantaneous feature extraction and normalization module 330 is used to: calculate the time change rate of each speech signal component to extract the instantaneous features corresponding to the speech signal components; adjust the amplitude value of the instantaneous features to the same preset range to obtain the current sub-signal corresponding to each of the speech signal components.

[0160] In one embodiment, the bandwidth optimization module 340 is used to: calculate the frequency distribution of the current sub-signal, evaluate the bandwidth of the current sub-signal according to the frequency distribution, and obtain an evaluation result; the evaluation result indicates whether the bandwidth meets the optimization target and the frequency distribution adjustment direction; according to the evaluation result, adjust the frequency distribution of the current sub-signal and the center frequency of the current sub-signal to obtain an updated current sub-signal; return to execute the step of calculating the frequency distribution of the current sub-signal until the evaluation result of the current sub-signal meets the optimization target, determine the current sub-signal as the target sub-signal, and obtain the center frequency corresponding to the target sub-signal.

[0161] In one embodiment, the periodic feature extraction module 350 is used to: perform orthogonal splitting of the target sub-signal corresponding to the center frequency to obtain a first split signal and a second split signal, perform periodic feature extraction on the first split signal and the second split signal to obtain multiple first periodic features of the first split signal and multiple second periodic features of the second split signal; perform combined analysis on the multiple first periodic features to obtain a periodic feature signal of the first oscillation mode, and perform combined analysis on the multiple second periodic features to obtain a periodic feature signal of the second oscillation mode; obtain a feature signal corresponding to each target sub-signal based on the periodic feature signal of the first oscillation mode and the periodic feature signal of the second oscillation mode corresponding to each target sub-signal.

[0162] In one embodiment, the speech classification and recognition module 360 ​​is used to: obtain the historical feature signal corresponding to the control command speech signal sample input to the surgical power device, and fuse the historical feature signal into a sample two-dimensional time-frequency diagram; the sample two-dimensional time-frequency diagram includes the periodic features and frequency features of the historical feature signal; the sample two-dimensional time-frequency diagram is input into the speech classification model to be trained, and the speech classification model extracts the local features and overall features of the sample two-dimensional time-frequency diagram, and outputs the command recognition result for the control command speech signal sample according to the local features and the overall features; according to the command recognition result of the control command speech signal sample, the model parameters of the speech classification model are adjusted, and when the training end conditions are met, the trained speech classification model is obtained.

[0163] In one embodiment, the speech classification and recognition module 360 ​​is used to: perform semantic understanding on the control command speech signal and convert the control command speech signal into a control command text; match the control command text with a plurality of command corpora in a pre-established command corpus to obtain a matching degree between the control command text and each of the plurality of command corpora; and update the command corpora in the command corpus according to each matching degree.

[0164] Each module in the above-mentioned speech recognition device for surgical power device can be implemented in whole or in part by software, hardware and their combination. Each module can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0165] In an exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG4. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected via a system bus, and the communication interface, the display unit, and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a speech recognition method for a surgical power device is implemented. The display unit of the computer device is used to form a visually visible picture, which may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0166] Those skilled in the art will understand that the structure shown in FIG. 4 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0167] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0168] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0169] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0170] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0171] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0172] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0173] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A speech recognition method for a surgical power device, characterized in that: The method comprises: Acquire a control command voice signal input for the surgical power device; Performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges; Extracting instantaneous features of each of the speech signal components, and normalizing the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components; Taking minimizing the bandwidth of each current sub-signal as the optimization goal, iteratively adjusting each current sub-signal, and determining each target sub-signal and the center frequency corresponding to each target sub-signal according to the iteration result; According to the center frequency of each of the target sub-signals, periodic feature extraction is performed on each of the target sub-signals to obtain a feature signal corresponding to each of the target sub-signals; Each of the feature signals is input into a trained speech classification model to obtain a control command recognition result output by the speech classification model.

2. The method according to claim 1, characterized in that The instantaneous feature extraction of each of the speech signal components and normalization of the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components includes: Calculating the time change rate of each of the speech signal components to extract the instantaneous features corresponding to the speech signal components; The amplitude values ​​of the instantaneous features are adjusted to the same preset range to obtain current sub-signals corresponding to the speech signal components.

3. The method according to claim 1, characterized in that: The step of iteratively adjusting each current sub-signal with minimizing the bandwidth of each current sub-signal as the optimization goal, and determining each target sub-signal and a center frequency corresponding to each target sub-signal according to the iteration result includes: Calculating the frequency distribution of the current sub-signal, and evaluating the bandwidth of the current sub-signal according to the frequency distribution to obtain an evaluation result; the evaluation result indicates whether the bandwidth meets the optimization goal and the frequency distribution adjustment direction; According to the evaluation result, adjusting the frequency distribution of the current sub-signal and the center frequency of the current sub-signal to obtain an updated current sub-signal; Return to the step of calculating the frequency distribution of the current sub-signal until the evaluation result of the current sub-signal meets the optimization target, determine the current sub-signal as the target sub-signal, and obtain the center frequency corresponding to the target sub-signal.

4. The method according to claim 1, characterized in that: The step of extracting periodic features of each target sub-signal according to the center frequency of each target sub-signal to obtain a feature signal corresponding to each target sub-signal includes: orthogonally splitting the target sub-signal corresponding to the center frequency to obtain a first split signal and a second split signal, and extracting periodic features from the first split signal and the second split signal to obtain a plurality of first periodic features of the first split signal and a plurality of second periodic features of the second split signal; Combining and analyzing the plurality of first periodic features to obtain a periodic characteristic signal of a first oscillation mode, and combining and analyzing the plurality of second periodic features to obtain a periodic characteristic signal of a second oscillation mode; The characteristic signal corresponding to each of the target sub-signals is obtained according to the periodic characteristic signal of the first oscillation mode and the periodic characteristic signal of the second oscillation mode corresponding to each of the target sub-signals.

5. The method according to claim 1, characterized in that The speech classification model is trained by the following steps, including: Acquire a historical feature signal corresponding to a sample of a control command voice signal input to a surgical power device, and fuse the historical feature signal into a sample two-dimensional time-frequency graph; the sample two-dimensional time-frequency graph includes a period feature and a frequency feature of the historical feature signal; Inputting the sample two-dimensional time-frequency graph into a speech classification model to be trained, extracting local features and overall features of the sample two-dimensional time-frequency graph by the speech classification model, and outputting a command recognition result for the control command speech signal sample according to the local features and the overall features; According to the command recognition result of the control command voice signal sample, the model parameters of the voice classification model are adjusted, and when the training end condition is met, the trained voice classification model is obtained.

6. The method according to claim 1, characterized in that After inputting each of the feature signals into a trained voice classification model to obtain a control command recognition result output by the voice classification model, the method further includes: Performing semantic understanding on the control command voice signal and converting the control command voice signal into a control command text; Matching the control command text with a plurality of command corpora in a pre-established command corpus to obtain a matching degree between the control command text and each of the plurality of command corpora; According to each of the matching degrees, the command corpus in the command corpus is updated.

7. The method according to claim 1, characterized in that The performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges includes: Decomposing the control command voice signal into a plurality of voice modal signals, and extracting timing information of each of the voice modal signals; Determining the center frequency of each of the speech modal signals according to the timing information; According to the center frequency, a frequency shift operation is performed on the speech mode signal to obtain the speech signal components under multiple frequency modes corresponding to different center frequency ranges.

8. A speech recognition device for a surgical power device, characterized in that: The device comprises: A voice signal acquisition module, used to obtain a control command voice signal input to the surgical power device; A signal decomposition and modulation module, used for performing signal decomposition and modulation on the control command voice signal to obtain voice signal components corresponding to the control command voice signal in frequency modes in different frequency ranges; An instantaneous feature extraction and normalization module extracts instantaneous features from each of the speech signal components, and normalizes the extracted instantaneous features of each of the speech signal components to obtain a current sub-signal corresponding to each of the speech signal components; A bandwidth optimization module, configured to iteratively adjust each of the current sub-signals with minimizing the bandwidth of each of the current sub-signals as the optimization target, and determine each of the target sub-signals and the center frequency corresponding to each of the target sub-signals according to the iteration result; A periodic feature extraction module, used to extract periodic features of each target sub-signal according to the center frequency of each target sub-signal, so as to obtain a feature signal corresponding to each target sub-signal; The speech classification and recognition module is used to input each of the feature signals into a trained speech classification model to obtain a control command recognition result output by the speech classification model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images