LED lamp structure capable of emitting light laterally in two directions

Abstract

The invention relates to the field of lighting devices, and discloses a bidirectional side light-emitting LED lamp structure which comprises a bonding pad, a support is installed on one side of the outer wall of the bonding pad, cavities are formed in the two sides of the outer wall of the support, LED lamp strips are installed in the cavities, a sound control system is installed on one side of the outer wall of the support, and the sound control system is connected with the LED lamp strips. The output end of the sound control system is electrically connected with the LED lamp strip. The sound control system comprises a sound collection module, a signal processing module and an instruction recognition module, the sound collection module is used for collecting sound signals, the signal processing module processes the collected sound signals, and the instruction recognition module recognizes sound instructions based on the processed signals. And the LED lamp strip executes corresponding operation according to the identification result. According to the invention, the technical scheme of sound signal preprocessing based on wavelet transform is adopted, so that the technical effects of effectively removing noise interference and accurately extracting local features of sound signals are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of lighting devices, and in particular to a bidirectional side-emitting LED lamp structure. Background Art

[0002] In the field of lighting and intelligent control, the voice control system of the bidirectional side-emitting LED lamp structure has broad application prospects, but the existing technology has many shortcomings. From the signal processing level, traditional voice control systems mostly rely on simple filtering algorithms and can only process stable signals. When faced with complex voice commands mixed with multi-frequency environmental noise, it is difficult to effectively remove noise interference, and it will also cover up the key features of the signal, making it impossible to accurately extract the local features of the sound signal, greatly affecting the accuracy of subsequent command recognition. For example, in a noisy factory environment, under the interference of complex noises such as the roar and friction of the machine, it is difficult for a simple filtering algorithm to separate the user's voice commands, making it difficult for the system to operate normally.

[0003] In terms of command recognition, existing voice control systems mostly use simple threshold judgment or fixed rule matching algorithms, which lack adaptability to environmental changes. Once the intensity or frequency of environmental noise changes, misjudgment is very likely to occur. Since it is impossible to establish an accurate probability model for voice commands, it becomes difficult to accurately identify voice commands in complex environments. In scenes like shopping malls where the background noise is complex and constantly changing, simple command recognition algorithms can easily misjudge background noise as commands, or fail to recognize the real commands masked by noise, causing the system to fail to respond normally.

[0004] In terms of testing and optimization mechanisms, many existing technologies only test voice control systems in simple environments and lack simulation of complex scenarios. Because various actual application scenarios are not fully considered, such tests cannot accurately reflect the performance of the system in a real environment. At the same time, due to the lack of quantitative evaluation indicators, it is difficult to scientifically evaluate the performance of the system, and it is impossible to optimize the system in a targeted manner. In actual applications, simple tests cannot discover potential problems of the system in complex environments, resulting in frequent failures of the system after it is put into use, affecting the user experience. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present invention provides a bidirectional side-emitting LED lamp structure, which solves the problem that the existing bidirectional LED lamp is difficult to effectively remove noise interference when facing complex sound commands mixed with multi-frequency environmental noise.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a bidirectional side-emitting LED lamp structure, including a soldering pad, a bracket is installed on one side of the outer wall of the soldering pad, cavities are opened on both sides of the outer wall of the bracket, an LED light strip is installed inside the cavity, a voice control system is installed on one side of the outer wall of the bracket, and the output end of the voice control system is electrically connected to the LED light strip.

[0007] Preferably, the voice control system includes a sound collection module, a signal processing module, and a command recognition module. The sound collection module is used to collect sound signals, the signal processing module processes the collected sound signals, the command recognition module recognizes sound commands based on the processed signals, and the LED light strip performs corresponding operations according to the recognition results.

[0008] Preferably, the signal processing module uses wavelet transform to process the sound signal, decomposing the sound signal into low-frequency approximate components and high-frequency detail components at different scales to extract local features of the sound signal. The wavelet transform uses Daubechies wavelet, and the order of the wavelet basis is determined through experiments to adapt to the characteristics and frequency range of the sound signal.

[0009] Preferably, the signal processing module processes the wavelet coefficients using a soft threshold denoising method, calculates the threshold according to Donoho's general threshold formula, removes noise interference, and retains effective features of the sound signal.

[0010] Preferably, the instruction recognition module classifies the voice instructions into multiple categories, extracts features from the denoised voice signal, and establishes a probability model of the voice instructions.

[0011] Preferably, the instruction recognition module extracts Mel-frequency cepstral coefficients as sound signal features, and obtains feature vectors through pre-emphasis, frame windowing, fast Fourier transform, Mel filter bank, logarithmic operation and discrete cosine transform steps.

[0012] Preferably, the probability model of the voice instructions assumes that the feature vector corresponding to each voice instruction obeys a Gaussian distribution, estimates the mean vector and the covariance matrix through a maximum likelihood estimation method, and determines the likelihood function.

[0013] Preferably, the instruction recognition module calculates the posterior probability based on Bayes' theorem, combined with the set prior probability and the calculated likelihood function, and selects the category with the largest posterior probability as the recognition result.

[0014] Preferably, the voice control system builds a simulated test environment, simulates different complex scenarios, generates test samples containing background noise, and uses indicators including accuracy, recall rate, and F1 value to evaluate system performance.

[0015] Preferably, the voice control system optimizes system performance according to the performance evaluation result and for voice command types with low recognition accuracy by selecting one of increasing training data to re-estimate probability model parameters or adjusting wavelet transform parameters.

[0016] The present invention provides a bidirectional side-emitting LED lamp structure, which has the following beneficial effects: 1. The present invention adopts a technical solution based on wavelet transform for sound signal preprocessing, which achieves the technical effect of effectively removing noise interference and accurately extracting local features of sound signals. Compared with the simple filtering denoising technical solution in the prior art, it solves the shortcomings of being unable to effectively process non-stationary sound signals and easily losing key features; 2. The present invention establishes a probability model of voice commands based on Bayesian inference and performs recognition, which achieves the effect of accurately recognizing voice commands in complex environments. Compared with the technical solutions in the prior art that rely on simple threshold judgment or fixed rule matching, it solves the problems of being difficult to adapt to changing environments and having low recognition accuracy; 3. The present invention builds a test environment that simulates a variety of complex scenarios, and combines the accuracy, recall rate, F1 value and other indicators to evaluate the system performance. This technical solution achieves a comprehensive and scientific evaluation of the voice control system performance. Compared with the existing technical solutions that only test in a simple environment or lack quantitative evaluation indicators, this solves the defects of being unable to accurately understand the real performance of the system and difficult to optimize in a targeted manner; 4. The present invention adopts a technical solution that flexibly adjusts training data and algorithm parameters according to the performance evaluation results, thereby achieving the purpose of continuously optimizing the performance of the voice control system, making the light evenly distributed on both sides, and avoiding the problem of light being concentrated in a certain area. Compared with the technical solution in the prior art that has fixed system parameters and is difficult to improve according to actual conditions, this solves the problem of being unable to adapt to different usage scenarios and changes in user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a three-dimensional diagram of the LED lamp structure of the present invention; Figure 2 This is a diagram showing the structure of the LED lamp of the present invention; Figure 3 An exploded view of the LED lamp structure of the present invention; Figure 4 This is a diagram of the architecture of the voice control system of the present invention.

[0018] Among them, 1. soldering pad; 2. bracket; 3. cavity; 4. LED light strip. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings of the present invention specification 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.

[0020] Please see attached Figure 1-4The embodiment of the present invention provides a bidirectional side-emitting LED lamp structure, including a pad 1, a bracket 2 is installed on one side of the outer wall of the pad 1, cavities 3 are opened on both sides of the outer wall of the bracket 2, an LED light strip 4 is installed inside the cavity 3, a voice control system is installed on one side of the outer wall of the bracket 2, and the output end of the voice control system is electrically connected to the LED light strip 4; The voice control system includes a voice collection module, a signal processing module, and a command recognition module. The voice collection module is used to collect voice signals, the signal processing module processes the collected voice signals, the command recognition module recognizes voice commands based on the processed signals, and the LED light strip performs corresponding operations according to the recognition results. The pad 1 is usually made of copper, a metal material with good electrical and thermal conductivity. As a key component connected to the external circuit, its size and shape are designed according to the power requirements and application scenarios of the LED lamp, providing a stable power supply for the LED light strip 4 and assisting in heat dissipation. The bracket 2 is made of insulating plastic or ceramic material with a certain mechanical strength to support and fix other components. One side of its outer wall is tightly connected to the pad 1 to ensure stable electrical connection. The cavity 3 opened on both sides of the outer wall is precisely designed according to the specifications of the LED light strip 4, which is used to place the light strip to ensure stable and accurate installation. The LED light strip 4 is composed of a plurality of closely arranged high-brightness, low-power LED lamp beads. When installed in the cavity 3, it is tightly fitted with the inner wall with thermal conductive glue, which can not only fix the light strip but also quickly conduct the heat generated by the work of the lamp beads to the bracket 2 for dissipation. The positive and negative poles of each lamp bead are connected to the pad 1 through a metal wire to form a complete circuit loop. The voice control system is the highlight of the invention, giving the LED lamp an intelligent interactive function. The sound collection module generally uses a high-sensitivity electret microphone, which is installed on the side of the bracket 2 close to the user's voice area. It works based on the principle of sound-to-electric conversion. A sound filter is also set at the front end. The filtering method is selected according to the noise characteristics of the actual scene. For example, a high-pass filter is used to filter out low-frequency noise in a noisy environment. After receiving the electrical signal from the sound collection module, the signal processing module performs a wavelet transform and selects the Daubechies wavelet. The optimal order is determined by experimentally comparing the signal feature clarity and noise residue after processing a variety of sound samples with wavelets of different orders. Then, the discrete wavelet transform is used to decompose the sound signal into low-frequency approximate components and high-frequency detail components of different scales. The Mallat algorithm is used for rapid calculation, and the soft threshold denoising method is used to process the wavelet coefficients to remove noise. Finally, the inverse transform is used to reconstruct the denoised sound signal. The command recognition module first classifies the sound commands, extracts the Mel-frequency cepstrum coefficients of the denoised sound signal, and obtains the MFCC feature vector through pre-emphasis, frame windowing, fast Fourier transform, Mel filter bank processing, logarithmic operation and discrete cosine transform. A probability model is established assuming that each command feature vector obeys a Gaussian distribution. The mean vector and covariance matrix are estimated using the maximum likelihood estimation method through a large amount of training data to determine the likelihood function. The posterior probability is calculated based on the Bayesian theorem combined with the prior probability and the likelihood function, and the category with the largest posterior probability is selected as the recognition result. The bidirectional side-emitting LED lamp structure has many beneficial effects. The bidirectional side-emitting design enables it to illuminate a larger area, reduce lighting blind spots, improve lighting efficiency, and is suitable for a variety of lighting scenarios; the voice control system allows users to conveniently control the switch, brightness and color mode of the LED lamp through sound, improving user experience and increasing the product's sense of technology and practicality; high-brightness, low-power lamp beads are selected and combined with an efficient heat dissipation design to achieve efficient lighting while reducing energy consumption, in line with the trend of energy conservation and environmental protection; the voice control system relies on advanced signal processing and command recognition technology to accurately identify user commands in complex environmental noise, ensuring stable operation and accurate response of the LED lamp.

[0021] In this embodiment, the sound collection module uses a microphone as a core component.

[0022] As an option, in some embodiments, a high-sensitivity electret microphone is used. The electret microphone has the characteristics of small size, low cost, and high sensitivity, and can better meet the sound collection requirements of the voice control system.

[0023] Specifically, when installing the electret microphone, its position layout needs to be considered. In one possible implementation, the electret microphone is set on the side of the LED lamp housing. In actual use scenarios, the side of the LED lamp is more likely to receive sounds from different directions in all directions. For example, when the user issues voice commands at various locations in the room, the microphone set on the side can more effectively capture the sound signal and reduce the occlusion and attenuation of the sound signal.

[0024] In order to further improve the effect of sound collection, some pre-processing measures can also be performed on the microphone. In some embodiments, a sound filter is set at the front end of the microphone. The filter can be designed as a high-pass filter, a low-pass filter or a band-pass filter according to actual needs. For example, when there is a lot of low-frequency noise interference in the environment, a high-pass filter can be used, and its cut-off frequency can be set to 500Hz. This can effectively filter out low-frequency noise below 500Hz, and only allow sound signals above 500Hz to pass, thereby improving the purity of sound collection.

[0025] In terms of signal transmission, the electrical signal collected from the microphone needs to be transmitted to the subsequent signal processing module. Generally, shielded cables are used for transmission to reduce the impact of external electromagnetic interference on the signal. The shielding layer of the shielded cable can effectively block external electromagnetic interference and ensure the integrity and accuracy of the sound signal during transmission.

[0026] In addition, in the extended technical content, different usage scenarios and sound collection requirements are taken into account. In some embodiments, microphone array technology can be used. By arranging multiple microphones in specific geometric shapes, such as linear arrays and circular arrays, directional collection and enhancement of sound can be achieved. For example, in a larger conference room scene, a linear microphone array can be used to accurately determine the direction of the sound source based on the arrival time difference and phase difference of the sound, and enhance the sound in that direction, thereby improving the effect and accuracy of sound collection and meeting the sound collection requirements in complex scenes.

[0027] The signal processing module uses wavelet transform to process the sound signal, decomposing the sound signal into low-frequency approximate components and high-frequency detail components at different scales to extract the local characteristics of the sound signal. The wavelet transform uses Daubechies wavelet, and the order of the wavelet basis is determined through experiments to adapt to the characteristics and frequency range of the sound signal. The signal processing module uses the soft threshold denoising method to process the wavelet coefficients, calculates the threshold according to the Donoho general threshold formula, removes noise interference, and retains the effective characteristics of the sound signal; In this embodiment, the signal processing module receives the electrical signal output from the sound collection module.

[0028] Generally speaking, the first operation performed by the signal processing module is wavelet transform.

[0029] As an option, in some embodiments, Daubechies wavelet is selected to perform wavelet transform. Daubechies wavelet has good time-frequency localization characteristics and can effectively analyze non-stationary signals such as sound.

[0030] Specifically, when determining the order of the Daubechies wavelet, rigorous experiments are required. Prepare Daubechies wavelets of different orders, such as Db2, Db4, and Db6. Select a variety of typical sound samples, such as daily conversation voices, environmental background sounds, and music clips. Use wavelets of different orders to transform and process these sample signals. Observe the feature clarity of the processed signal, such as the degree of retention of high-frequency details, the accuracy of low-frequency contours, and the residual noise. If the wavelet order is too low, the suppression effect on high-frequency noise is not good; if the order is too high, the signal may be over-smoothed, resulting in the loss of some useful features. After multiple rounds of comparative experiments, the wavelet order that is most suitable for the current voice control system is determined.

[0031] In one possible implementation, discrete wavelet transform (DWT) is used to decompose the sound signal, and the Mallat algorithm is used for fast calculation. The algorithm decomposes the signal through a set of conjugate image filters. Suppose the input sound signal is x[n]. After the j-th layer of decomposition, the low-frequency approximate component c j [n] and high-frequency detail component d j The calculation formula for [n] is as follows: Where h[n] and g[n] are the coefficients of the low-pass and high-pass filters respectively, j = 1, 2, ..., J, and J is the number of decomposition levels.

[0032] In some embodiments, after the decomposition is completed, the obtained wavelet coefficients are subjected to soft threshold denoising. The threshold T is calculated according to the Donoho general threshold formula: Where σ is the noise standard deviation, which is estimated by the median of the high-frequency detail components of the Jth layer, as follows: N is the signal length. The wavelet coefficients less than the threshold T are set to zero, and the coefficients greater than the threshold are retained. Then, the inverse discrete wavelet transform (IDWT) is performed to reconstruct the denoised sound signal.

[0033] In terms of extended technical content, in some embodiments, the characteristic differences of different environmental noises are taken into account. Before wavelet transform, a preliminary noise characteristic analysis can be performed on the sound signal. By statistically analyzing the frequency distribution and amplitude change parameters of the signal over a period of time, it is determined whether the type of noise is Gaussian white noise, impulse noise or other colored noise. According to the type of noise, the parameters of the wavelet transform, such as the number of decomposition layers and the selection of the wavelet basis, are adjusted in a targeted manner to further improve the denoising effect to adapt to more complex and changeable practical application environments.

[0034] The command recognition module classifies the voice commands into multiple categories, extracts features from the denoised voice signals, and builds a probability model for the voice commands. The instruction recognition module extracts the Mel frequency cepstral coefficients as the sound signal features, and obtains the feature vector through the steps of pre-emphasis, frame windowing, fast Fourier transform, Mel filter bank, logarithmic operation and discrete cosine transform; The probability model of the sound command assumes that the feature vector corresponding to each sound command obeys the Gaussian distribution, estimates the mean vector and covariance matrix through the maximum likelihood estimation method, and determines the likelihood function; The instruction recognition module calculates the posterior probability based on the Bayesian theorem, combining the set prior probability and the calculated likelihood function, and selects the category with the largest posterior probability as the recognition result; In the voice control system, after the signal processing module completes the preprocessing of the sound signal, the command recognition module takes on the key mission. It must accurately identify the voice commands from the processed signal to provide a basis for subsequent control of the LED lights.

[0035] In this embodiment, the instruction recognition module receives the de-noised and feature-extracted sound signal output by the signal processing module.

[0036] Generally speaking, the first task of the command recognition module is to classify the voice commands.

[0037] As an option, in some embodiments, the voice commands are divided into basic categories such as turning on and off the light, adjusting the brightness, and switching color modes. For example, if the user says "turn on the light", it is classified as the turning on and off light command category; if the user says "turn it brighter", it belongs to the brightness adjustment command category.

[0038] Specifically, the Mel Frequency Cepstral Coefficient (MFCC) features are extracted from the denoised signal. In a possible implementation, pre-emphasis is performed first. The pre-emphasis coefficient is generally set to 0.95, which is an empirical value. Using the formula: y[n]=x[n]-αx[n-1] Among them, x[n] is the input sound signal, y[n] is the pre-emphasized signal, and α is the pre-emphasis coefficient 0.95. This formula is used to increase the strength of the high-frequency signal.

[0039] Then, the frame is divided and windowed. In some embodiments, the length of each frame is set to 256 sampling points, and the frame shift is set to 128 sampling points. Using the Hamming window w[n], the formula is: n is the current sampling point number, N is the frame length 256, and windowing is performed on each frame of signal to avoid spectrum leakage.

[0040] Then, a fast Fourier transform (FFT) is performed. Through the FFT algorithm, the time domain signal is converted to the frequency domain to obtain the spectrum Y(k), which can clearly show the frequency components of the signal.

[0041] Then it passes through the Mel filter bank. The filter bank consists of multiple triangular filters, which maps the frequency domain signal to the Mel frequency scale, which conforms to the hearing characteristics of the human ear.

[0042] Then the mapped signal is logarithmically operated to enhance the characteristics of small amplitude signals and make them more obvious.

[0043] Finally, discrete cosine transform (DCT) is performed to obtain the MFCC feature vector. DCT can remove signal correlation and extract key features.

[0044] When building the probability model, it is assumed that the feature vector of each instruction follows a Gaussian distribution N(μ,∑). Where μ is the mean vector and ∑ is the covariance matrix. Using a large amount of training data, these two parameters are estimated by the maximum likelihood estimation method.

[0045] According to Bayes' theorem: The posterior probability P(θ|x) is calculated by combining the prior probability P(θ) and the likelihood function P(x|θ). The prior probability is set based on experience. For example, in daily use, the command "turn on the light" appears frequently, so its prior probability is set higher.

[0046] Select the category with the largest posterior probability to complete instruction recognition.

[0047] In terms of extended technical content, in some embodiments, the differences in voice characteristics of different users are taken into consideration. A personalized voice model can be established for each user. By collecting a large number of voice samples from each user and training a probability model separately, when recognizing instructions, the user's voice can be matched more accurately, the recognition accuracy can be improved, and the usage needs of different users can be met.

[0048] The voice control system builds a simulated test environment, simulates different complex scenarios, generates test samples containing background noise, and uses indicators including accuracy, recall rate, and F1 value to evaluate system performance; According to the performance evaluation results, the voice control system optimizes the system performance by increasing training data to re-evaluate probability model parameters or adjusting wavelet transform parameters for voice command types with low recognition accuracy. After the signal processing and command recognition modules of the voice control system are built and debugged, it is particularly important to build a simulated test environment to verify whether its performance meets the standards and whether it can operate stably in a real complex environment. It is a key link in evaluating system quality and discovering potential problems.

[0049] In this embodiment, the simulation test environment is set up to comprehensively test the performance of the voice control system.

[0050] Generally speaking, building a simulation test environment requires considering a variety of actual scenario factors.

[0051] As an option, in some embodiments, professional audio processing software is used to generate background noise.

[0052] Specifically, when simulating factory workshop scenes, audio processing software is used to generate composite background noise including machine roars and equipment operation sounds. The frequency range of these noises is wide, ranging from low-frequency machine vibrations to high-frequency friction sounds. Through software parameter settings, the intensity and frequency distribution of the noise can be adjusted, such as setting the low-frequency noise intensity to 60-80dB and the high-frequency noise intensity to 40-60dB to simulate the noisy environment of a real factory.

[0053] In a possible implementation, when simulating a shopping mall scene, in addition to crowd noise, background music is also added. The crowd noise can be recorded by real shopping mall audio and then processed by software as background noise. For background music, common shopping mall music is selected and mixed with crowd noise at a certain volume ratio. For example, the volume ratio of crowd noise to background music is set to 7:3, so that the simulated environment is closer to the actual shopping mall atmosphere.

[0054] In some embodiments, in order to simulate street scenes, background noise including car driving sounds, horn sounds, and pedestrian conversation sounds is generated. The frequency of car driving sounds varies with the speed of the car, and the sound frequencies at different speeds are simulated by software. For example, the frequency is 100-300Hz when driving at low speeds, and the frequency is 300-800Hz when driving at high speeds. In combination with the probability of occurrence of various sounds in actual streets, the background noise for testing is randomly combined.

[0055] After the background noise is built, the clean voice command signal is mixed with the background noise. The voice command signal includes the previously classified commands such as turning on and off the light, adjusting the brightness, and switching the color mode. The mixing is performed according to different signal-to-noise ratios (SNR). The signal-to-noise ratio calculation formula is: Among them, P signal is the power of the voice command signal, P noise is the power of the background noise. By adjusting P signal and P noise Set different signal-to-noise ratios, such as 5dB, 10dB, and 15dB, to simulate environments with different levels of noise.

[0056] In terms of extended technical content, in some embodiments, the spatial characteristics of sound are taken into account. The spatial positioning function of sound is added to the simulated test environment, and multiple speakers are placed in a certain layout to simulate the effect of sound coming from different directions. For example, a speaker array with a circular layout is used to enable the voice control system to process voice commands from different directions in the simulated environment, and more comprehensively test the performance of the system in a complex spatial acoustic environment.

[0057] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A bidirectional side-emitting LED lamp structure, characterized in that: The invention comprises a soldering pad (1), a bracket (2) being installed on one side of the outer wall of the soldering pad (1), cavities (3) being provided on both sides of the outer wall of the bracket (2), an LED light strip (4) being installed inside the cavity (3), and a voice control system being installed on one side of the outer wall of the bracket (2), the output end of the voice control system being electrically connected to the LED light strip (4).

2. The bidirectional side-emitting LED lamp structure according to claim 1, characterized in that: The voice control system includes a sound collection module, a signal processing module, and a command recognition module. The sound collection module is used to collect sound signals, the signal processing module processes the collected sound signals, the command recognition module recognizes sound commands based on the processed signals, and the LED light strip performs corresponding operations according to the recognition results.

3. The bidirectional side-emitting LED lamp structure according to claim 2, characterized in that: The signal processing module uses wavelet transform to process the sound signal, decomposing the sound signal into low-frequency approximate components and high-frequency detail components at different scales to extract the local characteristics of the sound signal. The wavelet transform uses Daubechies wavelet, and the order of the wavelet basis is determined through experiments to adapt to the characteristics and frequency range of the sound signal.

4. The bidirectional side-emitting LED lamp structure according to claim 3, characterized in that: The signal processing module uses a soft threshold denoising method to process the wavelet coefficients, calculates the threshold according to the Donoho general threshold formula, removes noise interference, and retains the effective features of the sound signal.

5. The bidirectional side-emitting LED lamp structure according to claim 2, characterized in that: The command recognition module classifies the voice commands into multiple categories, extracts features from the denoised voice signals, and establishes a probability model of the voice commands.

6. The bidirectional side-emitting LED lamp structure according to claim 5, characterized in that: The instruction recognition module extracts Mel frequency cepstral coefficients as sound signal features, and obtains feature vectors through pre-emphasis, frame windowing, fast Fourier transform, Mel filter bank, logarithmic operation and discrete cosine transform steps.

7. The bidirectional side-emitting LED lamp structure according to claim 5, characterized in that: The probability model of the voice command assumes that the feature vector corresponding to each voice command obeys Gaussian distribution, estimates the mean vector and covariance matrix through the maximum likelihood estimation method, and determines the likelihood function.

8. The bidirectional side-emitting LED lamp structure according to claim 5, characterized in that: The instruction recognition module calculates the posterior probability based on the Bayesian theorem, combined with the set prior probability and the calculated likelihood function, and selects the category with the largest posterior probability as the recognition result.

9. The bidirectional side-emitting LED lamp structure according to claim 2, characterized in that: The voice control system builds a simulated test environment, simulates different complex scenarios, generates test samples containing background noise, and uses indicators including accuracy, recall rate, and F1 value to evaluate system performance.

10. The bidirectional side-emitting LED lamp structure according to claim 9, characterized in that: According to the performance evaluation result, the voice control system optimizes the system performance by selecting one of increasing training data to re-evaluate probability model parameters or adjusting wavelet transform parameters for voice command types with low recognition accuracy.

Images