Noise reduction processing method and system for Bluetooth audio equipment

Abstract

The invention provides a noise reduction processing method and system for Bluetooth audio equipment, and the method comprises the steps: obtaining an audio signal, and generating spectrum distribution data based on the audio signal; according to the intensity value of each frequency band in the spectrum distribution data, extracting a frequency characteristic set of a multi-source interference component; for the frequency characteristic set, determining a dynamic fluctuation parameter corresponding to the frequency characteristic; adjusting the configuration of the filter according to the corresponding relation between the dynamic fluctuation parameter and the preset noise width; synchronously updating the time domain response parameter of the filter based on the amplitude change rate of the dynamic fluctuation parameter; calculating dynamic threshold upper and lower limits according to the time domain response parameters, and generating a dynamic threshold sequence; and performing gain compensation processing on the environment audio signal based on the dynamic threshold sequence to generate a noise reduction audio signal. Adaptive time domain response parameters and a dynamic threshold sequence are adopted, so that filter configuration can be adjusted in real time according to noise characteristics.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a noise reduction processing method and system for a Bluetooth audio device. Background Art

[0002] As a core research direction in the field of audio technology, noise reduction processing of Bluetooth audio devices is of decisive significance for improving user experience and achieving high-fidelity sound quality. With the widespread application of wireless Bluetooth audio devices, users' expectations for sound quality continue to increase, especially in complex and changeable noise environments. The quality of noise reduction technology directly affects the market competitiveness of the equipment. However, current noise reduction solutions mostly rely on fixed parameters or a single control strategy, which makes it difficult to cope with dynamically changing noise scenarios. This type of method performs well in static environments, but in actual use conditions with mixed noise or rapid changes, there are often problems with insufficient noise reduction or distorted sound quality, resulting in a decline in user experience.

[0003] The limitation of existing methods is that they lack the ability to adapt to real-time changes in environmental noise. Most traditional noise reduction systems use preset filter parameters or simple gain adjustments, and cannot flexibly respond according to the frequency distribution, intensity fluctuations or time domain characteristics of the noise. This rigid design is prone to imbalance in processing effects when faced with multi-source noise or sudden interference, either over-suppressing useful signals or failing to effectively weaken the target noise. The core challenge focuses on how to achieve dynamic adaptability of noise reduction parameters, especially the coordinated optimization of key technical factors such as the filter's Q value, time constant and threshold control. Since these factors have not been finely adjusted in real-time analysis, it is difficult for the system to adopt differentiated strategies for noise in different frequency bands, which in turn exposes the technical problem of insufficient noise reduction accuracy in complex environments. At the same time, the interaction between multiple parameters has not been fully considered, resulting in limited overall effectiveness of the control strategy. Therefore, how to analyze the characteristics of environmental noise in real time and synchronously adjust the multi-dimensional parameters such as filter Q value, time constant, threshold control and gain compensation to achieve accurate adaptation of the noise reduction algorithm in a mixed noise environment has become a key issue that needs to be solved in this study. Solving this problem will directly promote the improvement of the noise reduction capability of Bluetooth audio devices in dynamic scenes and meet the stringent requirements of high-fidelity audio transmission. Summary of the invention

[0004] The present invention provides a noise reduction processing method for a Bluetooth audio device, which mainly includes: The method comprises: obtaining an audio signal of the surrounding environment of the Bluetooth audio device, performing a fast Fourier transform operation on the audio signal, and generating spectrum distribution data; judging whether there is a multi-source interference component exceeding a preset threshold value according to the intensity value of each frequency band in the spectrum distribution data, and generating a frequency feature set of the multi-source interference component if so; calculating the rate of change of the intensity of each frequency band in the time domain window for the frequency feature set, and determining the corresponding dynamic fluctuation parameter; adjusting the filter configuration according to the corresponding relationship between the dynamic fluctuation parameter and the preset noise width; synchronously updating the time domain response parameter of the filter based on the amplitude change rate of the dynamic fluctuation parameter; calculating the upper and lower limits of the dynamic threshold sequence according to the time domain response parameter, and triggering a threshold increment adjustment operation and updating the dynamic threshold sequence when it is detected that the spectrum distribution data exceeds the upper limit of the dynamic threshold sequence; performing gain compensation processing on the audio signal based on the updated dynamic threshold sequence to generate a noise reduction audio signal; detecting the residual noise intensity of the noise reduction audio signal, and if the residual noise intensity exceeds the set standard, repeatedly performing the filter configuration adjustment and the time domain response parameter update operation until the noise reduction requirement is met.

[0005] The technical solution provided by the embodiment of the present invention may include the following beneficial effects: by performing fast Fourier transform, spectrum analysis and dynamic fluctuation parameter extraction on the audio signal of the surrounding environment of the Bluetooth audio device, accurate identification of the complex noise environment is achieved. The innovative use of adaptive time constant sets and dynamic threshold sequences enables the filter configuration to be adjusted in real time according to the noise characteristics. Through gain compensation processing and residual noise detection, the present invention continuously optimizes the noise reduction effect. Especially in complex scenarios such as multi-source interference and dynamic frequency changes, this method can effectively suppress various types of noise and improve the clarity of audio signals. The core of the present invention lies in its adaptability and real-time performance, which can quickly respond to environmental changes, minimize noise interference while ensuring sound quality, and significantly improve the audio processing effect and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 The present invention is a flowchart of a noise reduction processing method for a Bluetooth audio device.

[0007] Figure 2 It is a detailed flow chart of step S103 of the present invention.

[0008] Figure 3 It is a detailed flow chart of step S104 of the present invention. DETAILED DESCRIPTION

[0009] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than 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.

[0010] See also Figure 1 In this embodiment, a noise reduction processing method for a Bluetooth audio device may specifically include: S101: Acquire an audio signal from the surrounding environment of the Bluetooth audio device, perform a fast Fourier transform operation on the audio signal, and generate spectrum distribution data.

[0011] Acquire the audio signal in the environment, capture the original waveform through the sensor, and obtain the initial audio data. Perform preprocessing operations on the initial audio data, use filtering technology to remove noise, and obtain purified audio data. Perform fast Fourier transform on the purified audio data to generate corresponding spectrum distribution data. Extract frequency components based on the spectrum distribution data, and determine the main frequency range by calculation. If the main frequency range exceeds the preset threshold, smooth the spectrum distribution data to obtain optimized spectrum data. Analyze the real-time change trend based on the optimized spectrum data to determine the dynamic characteristics of the audio signal. Use a pre-established machine learning classification model to perform pattern recognition on the dynamic characteristics and determine the category characteristics of the environmental audio.

[0012] Specifically, this scheme describes a method for audio signal processing and classification. First, the original audio waveform is captured by a sensor to obtain initial audio data.

[0013] Exemplarily, a microphone array may be used to collect ambient sound to form a multi-channel audio signal.

[0014] In a possible implementation manner, the initial audio data is preprocessed.

[0015] Specifically, a bandpass filter can be used to remove low-frequency environmental noise and high-frequency interference to improve the signal-to-noise ratio. For example, a bandpass filter with a cutoff frequency of 100Hz-8kHz can effectively retain valid signals such as human voice.

[0016] It should be noted that by performing a fast Fourier transform (FFT) on the preprocessed audio data, a frequency spectrum distribution in the frequency domain can be obtained.

[0017] In one embodiment, a 4096-point FFT is selected to obtain a higher frequency resolution. The main frequency components are extracted by analyzing the spectrum. For example, the frequency band range where the energy is concentrated is calculated, and the frequency interval where 80% of the energy is located is determined as the main frequency range.

[0018] Preferably, if the main frequency range exceeds a preset threshold, such as 4 kHz, the spectrum data is smoothed.

[0019] It can be understood that smoothing helps to suppress the impact of burst noise.

[0020] In one embodiment, the moving average method is used to smooth the spectrum, and the window size can be selected to be 5-10 frequency points. According to the optimized spectrum data, the dynamic characteristics of the audio signal are analyzed. For example, time domain features such as short-time energy and zero-crossing rate, as well as frequency domain features such as spectrum centroid and spectrum flux are calculated to comprehensively reflect the time-frequency characteristics of the audio. These features can serve as an important basis for audio classification. Finally, the audio is classified using a pre-trained machine learning model. In one possible implementation, a support vector machine (SVM) classifier is used, a feature vector is input, and an audio category is output. Through this method, different types of sounds such as speech, music, and traffic noise in the environment can be effectively identified, providing a basis for subsequent intelligent processing and decision-making.

[0021] S102: judging whether there is a multi-source interference component exceeding a preset threshold value according to the intensity value of each frequency band in the spectrum distribution data, and if so, generating a frequency feature set of the multi-source interference component.

[0022] The intensity value of each frequency band is obtained through the spectrum distribution data, and the preset threshold is used to determine whether there is a signal strength exceeding the threshold, so as to obtain the preliminary identification result of multi-source interference. Based on the preliminary identification result, the fast Fourier transform is used to extract the frequency characteristics of the interference component and generate the corresponding feature set. According to the feature set, the preset threshold is used for screening to determine the distribution data of multi-source interference. Through the distribution data, the time series analysis method is used to analyze the changing trend of the spectrum distribution and determine the dynamic characteristics of the interference component. According to the dynamic characteristics, the support vector machine algorithm is used for pattern classification to obtain the category characteristics of the interference component. According to the category characteristics, the fluctuation range of the signal strength is extracted, and the frequency band division algorithm is used to determine the frequency band division of multi-source interference. Through the frequency band division, the boundary conditions of the feature set are adjusted to obtain the optimized frequency characteristics.

[0023] Specifically, when obtaining the intensity value of each frequency band through the spectrum distribution data, the spectrum can be divided into several sub-bands, such as a low frequency band of 100 Hz-1 kHz, a middle frequency band of 1 kHz-4 kHz, and a high frequency band of 4 kHz-8 kHz.

[0024] For example, if the intensity value of a certain frequency band reaches -20dB after the collected ambient audio signal is transformed by fast Fourier transform, and the preset threshold is -30dB, it means that the signal intensity in this frequency band exceeds the threshold. This may be caused by external interference from multiple sources, such as wind or mechanical vibration.

[0025] In a possible implementation, when the strength value exceeds the standard, the frequency band can be preliminarily marked as a potential interference area to provide a basic basis for subsequent analysis.

[0026] It should be noted that when using fast Fourier transform to extract the frequency characteristics of interference components, we can focus on the frequency band that exceeds the threshold. For example, when selecting a 2048-point FFT to analyze a certain audio segment, we find that a significant peak appears near 2kHz, which may correspond to a periodic interference source, such as low-frequency resonance when the air conditioner is running. When generating a feature set, information such as peak frequency, amplitude, and bandwidth can be recorded to form a multi-dimensional feature vector. This method helps to clearly characterize the frequency characteristics of the interference.

[0027] Specifically, when a preset threshold is used for screening according to a feature set, the amplitude threshold may be set to -25 dB, and only frequency components above this value are retained.

[0028] In one embodiment, after screening, it is found that the features are more concentrated in the 1.5kHz-2.5kHz interval, indicating that there may be stable interference distribution data here. This step can effectively eliminate irrelevant noise and improve the accuracy of subsequent analysis. When performing time series analysis through distribution data, the trend of spectrum distribution over time can be observed. For example, in a 10-second audio clip, if the intensity value near 2kHz shows a periodic enhancement, it may suggest that the interference source has dynamic characteristics, such as sound fluctuations when a vehicle passes by.

[0029] Preferably, the rate of change of the intensity value can be calculated to determine the persistence or instantaneity of the interference, providing a dynamic basis for classification.

[0030] In one embodiment, when using a support vector machine algorithm to classify the pattern of interference components, a feature vector containing a frequency peak and a time rate of change may be input. For example, after training the model, it is identified that 2kHz interference belongs to mechanical noise, while 6kHz high-frequency interference may come from electronic equipment howling. This classification result helps to distinguish the source of interference and improve the precision of identification. When extracting the fluctuation range of signal strength based on category features, the amplitude change of a certain type of interference within 5 seconds can be analyzed.

[0031] For example, the intensity of mechanical noise fluctuates between -22dB and -18dB. When using the frequency band division algorithm to further determine the boundaries of multi-source interference, the frequency band with a larger fluctuation range can be divided separately, such as dividing 1.8kHz-2.2kHz into an independent interval. This detailed division can more accurately reflect the distribution of interference.

[0032] It is understandable that when adjusting the boundary conditions of the feature set, if it is found that the frequency band boundary of the interference near 2kHz is not clear enough, the feature value can be recalculated by expanding the analysis window to 3kHz. For example, after the feature originally concentrated at 2kHz is expanded to 2.5kHz, the frequency characteristics of the interference are more complete. This optimization can significantly improve the robustness of the feature set and provide more reliable data support for subsequent processing.

[0033] The detailed flow chart of step S103 of the present invention can be found in Figure 2 ,like Figure 2 As shown, S103, for the frequency feature set, calculate the change rate of each frequency band intensity in the time domain window, and determine the corresponding dynamic fluctuation parameter.

[0034] The original data of frequency band intensity is obtained from the feature set, and a fixed-length sliding window division method is used to generate a time domain window sequence to obtain sequence data; for the sequence data, a first-order difference method is used to calculate the change of continuous values ​​of the intensity of each frequency band to obtain the change rate; a preset threshold is obtained to determine whether the change rate exceeds the preset threshold range, and if it exceeds the preset threshold range, the corresponding change rate is marked as a significant fluctuation to determine the dynamic fluctuation range; for the fluctuation mode within the dynamic fluctuation range, a K-means clustering algorithm is used for grouping to obtain an intensity distribution category; the frequency band intensity fluctuation characteristics of each intensity distribution category are extracted from the time domain window sequence to determine the boundary adjustment parameters of the feature set; based on the adjusted feature set, Fourier transform is used to analyze the trend of the sequence data, determine the periodic characteristics of the dynamic fluctuation, and obtain the periodic parameters; based on the periodic parameters, the mean and variance of the intensity distribution are calculated by statistical methods to determine the final dynamic fluctuation parameters.

[0035] Specifically, when the raw data of frequency band intensity is extracted from the feature set, it can be regarded as a continuous sequence of audio signal intensity.

[0036] For example, suppose a piece of ambient audio is collected, the spectrum range is 100Hz-8kHz, and the intensity value is recorded in decibels. When using the fixed-length sliding window division method to generate a time domain window sequence, the window length can be set to 1 second and the step length to 0.5 seconds to generate multiple overlapping windows.

[0037] For example, for 10 seconds of audio, 19 windows can be obtained, each containing frequency band intensity data in the corresponding time period. This division method helps to capture the trend of intensity changes over time. When the first-order difference method is used to calculate the rate of change for sequence data, the principle is to reflect continuous changes through the difference between the intensity values ​​of adjacent windows.

[0038] Specifically, assuming that the intensity of a window in the 2kHz frequency band is -22dB, and the intensity of the frequency band of the next window is -20dB, the frequency band amplitude change rate is 2dB / 0.5 seconds, that is, 4dB / second.

[0039] In one possible implementation, the change rate threshold is set to 3dB / s, and if it exceeds the threshold, it is marked as a significant fluctuation. For example, if 4dB / s exceeds the threshold, it indicates that the frequency band may be affected by external dynamic interference, such as sudden mechanical impact. This method can quickly identify sudden changes in intensity and provide clues for subsequent analysis.

[0040] When using the K-means clustering algorithm to group the dynamic fluctuation range, the change rate can be used as an input feature. Preferably, the number of clusters K is set to 3, representing low, medium, and high fluctuation modes respectively.

[0041] For example, the clustering results may show that the fluctuations around 2kHz are mostly concentrated in the high fluctuation group, while those around 6kHz belong to the low fluctuation group. This grouping can preliminarily distinguish the dynamic characteristics of the interference. When extracting the fluctuation characteristics of the intensity distribution category from the time domain window, we can focus on the high-frequency or low-frequency proportion of each category.

[0042] In one embodiment, it is found that the frequency band of the high fluctuation group is concentrated in 1.8kHz-2.2kHz, indicating that there may be significant dynamic interference sources here. When determining the boundary adjustment parameters of the feature set, the frequency band range can be adjusted according to the clustering results. For example, the original boundary is around 2kHz, and after adjustment it is extended to 2.5kHz to cover more fluctuation features. When using Fourier transform to analyze the periodicity of sequence data, the principle is to convert the time domain window into the frequency domain and look for periodic peaks.

[0043] Specifically, by transforming the 10-second sequence, it is possible to find an intensity increase that repeats every 2 seconds near 2kHz, suggesting that the interference source has periodic characteristics, such as the sound of a fan. The period parameter can be recorded as 2 seconds. When the mean and variance of the intensity distribution are calculated using the period parameter, the statistical method can further quantify the dynamic fluctuations. For example, in the 2kHz frequency band, the mean intensity within 10 seconds is -21dB and the variance is 2dB, indicating that the fluctuation is relatively stable, where the mean represents the average level of dynamic fluctuations and the variance represents the degree of dispersion of dynamic fluctuations.

[0044] It is understandable that this quantification result helps to determine the persistence of interference. In one embodiment, if the variance is small, the interference may be stable background noise; if the variance is large, it may be an instantaneous impact sound. This analysis method can provide a reliable basis for interference classification and subsequent processing, while improving the comprehensiveness of feature description.

[0045] Based on the results of the above steps, the final dynamic fluctuation parameters are determined. These dynamic fluctuation parameters may include the amplitude change rate, mean, variance, etc. of the fluctuation for further analysis or application. The detailed flow chart of step S104 of the present invention can be found in Figure 3 ,like Figure 3 As shown, S104, adjusting and generating a filter configuration according to the corresponding relationship between the dynamic fluctuation parameter and the preset noise width; The dynamic fluctuation parameter and the noise width value are obtained, and the adjustment coefficient corresponding to the dynamic fluctuation parameter and the noise width value is searched from the parameter mapping table; for the preset frequency band, an adjustment sequence corresponding to each frequency band is generated according to the adjustment coefficient; if the adjustment sequence exceeds the preset range, the adjustment sequence is corrected by a linear interpolation method to obtain a corrected adjustment sequence; the initial configuration parameters of the filter are generated according to the corrected adjustment sequence, and the configuration adjustment value is calculated according to the matching degree between the initial configuration parameters and the dynamic fluctuation, and the initial configuration parameters are optimized by the configuration adjustment value to obtain an optimized filter configuration; it is determined whether the optimized filter configuration satisfies the preset relationship, and if so, the optimized filter configuration is output, and if not, the optimized filter configuration is iteratively optimized by a gradient descent method.

[0046] Specifically, the acquisition of dynamic fluctuation parameters and noise width values ​​is targeted. It can be understood that this step is intended to extract key parameters from the signal to reflect its changing characteristics.

[0047] Exemplarily, assuming that the dynamic fluctuation parameter of an audio signal is defined as the rate of amplitude change, the noise width value represents the range of the interference frequency band.

[0048] In a possible implementation, a section of environmental sound is collected, and the dynamic fluctuation parameter may be recorded as 5dB / second and the noise width value may be 200Hz. These values ​​can be obtained from time domain or frequency domain analysis.

[0049] Specifically, the fluctuation value is determined by segmenting the signal and calculating the intensity difference between windows, while the noise width identifies the frequency bandwidth of the interference through spectrum analysis.

[0050] It should be noted that the mapping table is usually established in advance based on experimental data, reflecting the relationship between the fluctuation value and the noise width and the adjustment requirements. For example, a fluctuation value of 5dB / s and a noise width of 200Hz may correspond to an adjustment coefficient of 1.2. This coefficient is designed to amplify or reduce the amplitude of subsequent frequency band adjustments. In one embodiment, if the fluctuation value is higher, the coefficient may be larger to more strongly suppress interference.

[0051] Preferably, the intensity value of each frequency band is adjusted proportionally according to the adjustment coefficient.

[0052] Specifically, assuming the frequency band range is 1kHz to 3kHz, after the adjustment coefficient 1.2 is applied, the frequency band with the original strength value of -30dB may become -36dB. This adjustment sequence reflects the initial optimization intention for the signal. If the adjustment sequence exceeds the preset range, for example, the strength value exceeds the minimum threshold of -40dB, the linear interpolation method is used for correction.

[0053] It can be understood that linear interpolation generates reasonable correction values ​​by smoothly transitioning between the over-limit point and the adjacent point, such as interpolating -42dB to -40dB, to ensure that the sequence meets expectations. When generating the initial configuration parameters of the filter according to the corrected adjustment sequence.

[0054] In one possible implementation, the parameters may include the filter's cutoff frequency and gain. For example, if the correction sequence shows a sudden change in intensity at 1.5kHz, the initial configuration may set the cutoff frequency to 1.6kHz. The matching degree is calculated by comparing the degree of fit between the configuration parameters and the dynamic fluctuations.

[0055] For example, if the matching degree is 85%, the configuration adjustment value may be 0.1, which is used to fine-tune the filter cutoff frequency to 1.61kHz. This optimized configuration can filter out interference more specifically. When determining whether the optimized filter configuration satisfies the preset relationship.

[0056] Specifically, the preset relationship may be that the noise in the frequency band is reduced to below -35dB. If it is not satisfied, for example, a certain frequency band is still -33dB, the gradient descent method is used for iterative optimization.

[0057] In one embodiment, gradient descent gradually adjusts parameters such as gain or bandwidth to gradually approach the target configuration. This method can effectively improve the adaptability of the filter and ensure its practicality in a dynamic environment. For example, after optimization, the noise drops to -36dB, indicating that the interference is effectively suppressed. The advantage of this method is that it gradually approaches the optimal solution and is suitable for complex signal processing scenarios.

[0058] S105. Synchronously update the time domain response parameters of the filter based on the amplitude change rate of the dynamic fluctuation parameter.

[0059] The instantaneous value of the amplitude change is obtained, the distribution characteristics of the change rate are determined, and a preliminary adjustment strategy for the time domain response is generated based on the distribution characteristics; a quantization sequence of the adjustment requirements is obtained according to the preliminary adjustment strategy, and the deviation value of the response parameter is determined using the quantization sequence to obtain an initial mapping table of time constants; the time constant set is optimized through the initial mapping table, and the filter configuration parameters are updated using the optimized time constant set to obtain the final time domain response configuration.

[0060] Specifically, obtaining the instantaneous value of the amplitude change is an important step in analyzing the dynamic characteristics of the signal.

[0061] It can be understood that the instantaneous value of the amplitude change reflects the amplitude change of the signal at a certain moment.

[0062] For example, in an audio signal, assuming that the amplitude at a certain moment jumps from -20dB to -15dB through sampling, the instantaneous value can be defined as the difference of this change, that is, 5dB. To determine the distribution characteristics of the change rate, it is necessary to statistically analyze the change trend of the instantaneous value over a period of time.

[0063] In one possible implementation, signal samples are collected within 1 second, and the rate of each amplitude jump is recorded, such as 5dB / 0.1 second, 3dB / 0.2 second, etc. After drawing a distribution graph, it is found that the change rate is mostly concentrated in the 4-6dB / second range.

[0064] Preferably, if the rate concentration is high, a smoothing process may be preferred to reduce the impact of mutations. For example, the adjustment strategy may be initially determined to reduce the response spike. A quantitative sequence of adjustment requirements is obtained according to the preliminary adjustment strategy.

[0065] Specifically, the quantitative sequence is to convert the adjustment intention into an operational numerical sequence. Assuming that the distribution characteristics show a high rate of change, the quantitative sequence may be designed as a smoothing coefficient sequence, such as 0.8, 0.9, and 1.0, which correspond to the response adjustment amplitude in different time periods. When the quantitative sequence is used to determine the deviation value of the response parameter.

[0066] In one embodiment, assuming that the ideal response parameter is smoothness 1.0 and the actual value is 1.2, the deviation value is 0.2. The initial mapping table of time constants is obtained and a mapping relationship is established based on these deviation values. For example, if the deviation value 0.2 corresponds to the time constant adjusted to 0.01 seconds, the mapping table may record a set of corresponding relationships between deviations and time constants, such as 0.1 corresponding to 0.005 seconds and 0.2 corresponding to 0.01 seconds. This mapping table provides a basic basis for subsequent optimization.

[0067] It should be noted that the optimization goal is to make the time constant more adaptable to the dynamic characteristics of the signal.

[0068] In a possible implementation, assuming that the initial time constant set is 0.01 seconds, 0.02 seconds, and 0.03 seconds, after adjustment according to the mapping table, it may become 0.009 seconds, 0.018 seconds, and 0.025 seconds to more precisely match the change rate distribution.

[0069] Specifically, the time constant directly affects the response speed of the filter. For example, adjusting the time constant from 0.01 seconds to 0.009 seconds may make the filter more sensitive to rapidly changing amplitudes. After obtaining the final time domain response configuration, the filter can better adapt to amplitude jumps while maintaining signal integrity. Consider from multiple aspects. For example, the acquisition of instantaneous values ​​can be combined with different sampling rates to verify its accuracy. For example, increasing the sampling rate to 10kHz may reveal more subtle amplitude changes, such as 3dB / 0.05 seconds, making the distribution characteristics more accurate. The design of the quantization sequence can also be changed from smoothing to enhancement. Assuming that the signal needs to highlight the changing characteristics, the sequence can be adjusted to 1.1, 1.2, and 1.3. This flexibility ensures the diversity of solutions.

[0070] Preferably, the optimized time constant can also reduce the delay of the filter in high dynamic scenes and improve real-time performance.

[0071] It can be understood that this method enhances the filter's ability to process complex signals by gradually adjusting parameters, and has significant advantages especially in dynamic environments.

[0072] S106 , calculating the upper and lower limits of a dynamic threshold sequence according to the time domain response parameter, and when it is detected that the spectrum distribution data exceeds the upper limit of the dynamic threshold sequence, triggering a threshold increment adjustment operation and updating the dynamic threshold sequence.

[0073] Acquire the time domain response and calculate the upper and lower limits of the dynamic threshold to determine the detection range of the spectrum distribution data. If the spectrum distribution data exceeds the upper limit, the threshold increment adjustment is triggered to obtain the adjusted threshold increment. The dynamic threshold is updated using the adjusted threshold increment to obtain a new upper and lower limit. Determine whether the adjustment needs to be triggered again based on the new upper and lower limits. For the threshold sequence in the stable state, cluster the distribution data using the K-means algorithm. Optimize the trigger conditions based on the clustering results to obtain the final dynamic threshold sequence configuration.

[0074] Specifically, obtaining the time domain response is one of the basic steps to analyze the dynamic characteristics of the signal, which reflects the real-time reaction of the system to the input signal.

[0075] For example, when processing an audio signal, time domain response data can be obtained by sampling. Assume that the signal amplitude changes from -10dB to -8dB at a certain moment, and the duration is 0.2 seconds. This change trend provides a basis for the subsequent calculation of the dynamic threshold. When calculating the upper and lower limits of the dynamic threshold.

[0076] It will be appreciated that the upper and lower limits define the range within which the signal normally fluctuates.

[0077] In one possible implementation, the initial upper limit is set to -5dB and the lower limit is set to -15dB, which is based on the historical data statistics of the time domain response. If the spectrum distribution data exceeds the upper limit, for example, the amplitude detected in a certain sampling is -4dB, the threshold increment adjustment is triggered.

[0078] Specifically, the incremental adjustment may be to increase the upper limit by 0.5 dB to obtain a new upper limit of -4.5 dB, while the lower limit remains unchanged. When the dynamic threshold is updated using the adjusted threshold increment.

[0079] It should be noted that the new upper and lower limits will affect the sensitivity of subsequent tests.

[0080] In one embodiment, the updated upper limit is -4.5dB, and the lower limit is still -15dB. If the next sampling value still exceeds the upper limit, for example, reaches -3dB, the adjustment is triggered again, and the upper limit may become -3.5dB. This iterative method ensures the dynamic adaptability of the threshold. When determining whether to adjust again based on the new upper and lower limits.

[0081] Preferably, a stable condition can be set, for example, if five consecutive sampling values ​​are within the range, the current threshold is considered applicable. For example, if the consecutive sampling values ​​are -6dB, -7dB, -8dB, -9dB, and -10dB, all between -3.5dB and -15dB, no further adjustment is required.

[0082] Specifically, assuming that 100 sampling points are collected and the amplitude is distributed between -15dB and -5dB, the K-means algorithm divides them into three categories, with the center points being -12dB, -9dB, and -6dB respectively.

[0083] In one possible implementation, based on the clustering results, it is found that the points near -6dB are relatively dense, which may indicate that the signal fluctuates frequently in this range. Therefore, the trigger condition can be adjusted to pay more attention to this interval, such as tightening the upper limit to -5.5dB. Preferably, this optimization can capture the dynamic changes of the signal more accurately. From multiple aspects, the sampling rate for obtaining the time domain response will affect the delicacy of the data. For example, increasing it to 5kHz may capture a small jump from -10dB to -9dB within 0.1 seconds, making the threshold calculation more accurate. When calculating the upper and lower limits, if the initial range is adjusted in combination with the signal type, such as relaxing the low-frequency signal to -20dB to 0dB, it will be more applicable. The number of categories of K-means clustering can also be adjusted according to the complexity of the data. For example, increasing it to 5 categories can further refine the distribution characteristics. This multi-angle verification and adjustment ensures the adaptability and stability of the solution in a single scenario, while improving the efficiency of real-time detection.

[0084] S107: Perform gain compensation processing on the audio signal based on the updated dynamic threshold sequence to generate a noise reduction audio signal.

[0085] Acquire the ambient audio signal, and obtain the initial audio data through spectral feature extraction. For the initial audio data, use a preset dynamic threshold sequence to determine the noise compensation range and obtain the corresponding gain adjustment parameter. If the spectral feature exceeds the upper threshold, the gain adjustment process is triggered to obtain the adjusted audio signal. According to the adjusted audio signal, perform gain compensation to generate a preliminary noise reduction signal. Optimize the threshold sequence through a clustering algorithm, use the optimized threshold sequence to update the compensation range, and obtain the final noise reduction audio signal.

[0086] Specifically, obtaining environmental audio signals is the first step in analyzing audio characteristics, which directly reflects the acoustic information in the environment. For example, in a noisy cafe, the signal captured by the microphone may contain human voices, background music, and the sound of cups and plates colliding. These signals can be processed by fast Fourier transform to extract spectral features and obtain initial audio data.

[0087] Specifically, assuming that the sampling rate is set to 4kHz, after collecting a 1-second signal, the spectrum shows that the low-frequency band is concentrated around 100Hz, with an amplitude of about -12dB, while the high-frequency band, such as 2kHz, has an amplitude of -20dB. This spectrum distribution provides a basis for subsequent processing. For the initial audio data, it is a common strategy to use a preset dynamic threshold sequence to determine the noise compensation range. Exemplarily, the initial threshold sequence is set to an upper limit of -10dB and a lower limit of -25dB.

[0088] In a possible implementation, if the low frequency band -12dB is within the range, no adjustment is required; but if the high frequency band of a certain sampling jumps to -8dB, exceeding the upper limit, it indicates that the noise may be enhanced.

[0089] It should be noted that the gain adjustment parameters can be set accordingly, for example, applying a -2dB gain suppression to the high frequency band to balance the noise impact.

[0090] It can be understood that this method can effectively distinguish between signals and noise. If the spectrum feature exceeds the upper threshold, the gain adjustment process is triggered.

[0091] Preferably, the adjusted audio signal will be closer to the expected characteristics. For example, if the amplitude at 500Hz in a certain sampling rises to -9dB, exceeding the -10dB upper limit, the gain can be reduced by 1dB to make it fall back to -10dB.

[0092] In one embodiment, this adjustment can reduce sudden noise interference and ensure the stability of the audio. Performing gain compensation based on the adjusted audio signal is the key to generating a preliminary noise reduction signal. For example, +1dB compensation is applied to the low-frequency band to enhance the clarity of the human voice, while maintaining suppression on the high-frequency band. This differentiated processing can highlight the main audio content. Specifically, if the human voice in the cafe is concentrated at 300Hz, its amplitude will be increased from -12dB to -11dB after compensation, which sounds more natural. Optimizing the threshold sequence through a clustering algorithm can further improve the adaptability of the solution.

[0093] In one possible implementation, 200 sampling points are collected, with amplitudes ranging from -25dB to -5dB, and the K-means algorithm is used to classify them into three categories, with the center points being -20dB, -15dB, and -8dB. It is found that the points near -8dB are densely populated, which may reflect the frequent occurrence of high-frequency noise, so the upper limit is adjusted to -9dB and the lower limit is tightened to -22dB.

[0094] Preferably, this optimization can lock the noise range more accurately. After updating the compensation range with the optimized threshold sequence, the final noise-reduced frequency signal is obtained. For example, after the new threshold is applied, the amplitude at 500Hz is stabilized within -10dB, and the high-frequency noise is effectively suppressed.

[0095] In one embodiment, increasing the sampling rate to 8 kHz can capture more subtle spectrum jumps, such as a change from -15 dB to -14 dB within 0.05 seconds, making compensation more timely.

[0096] Understandably, this improves the real-time performance of noise reduction. From multiple perspectives, if the initial threshold range is relaxed to -30dB to 0dB, it will be applicable to more complex environments; if the number of clusters is increased to 5, the distribution characteristics will be more refined, which together support the robustness of the solution.

[0097] S108, detecting the residual noise intensity of the noise reduction audio signal, and if the residual noise intensity exceeds a set standard, repeatedly performing the filter configuration adjustment and the time domain response parameter update operations until the noise reduction requirement is met.

[0098] Obtain the noise spectrum of the noise reduction audio signal, determine the residual noise distribution through spectrum analysis, and obtain the residual noise intensity. If the residual noise intensity exceeds the preset threshold, use the adjustment tool to update the Q value parameter to obtain a new parameter configuration. For the optimized time domain response data, determine whether the residual noise intensity meets the preset noise reduction standard. If not, repeat the Q value parameter adjustment and time domain response update until the noise reduction standard is met.

[0099] Specifically, obtaining the noise spectrum of the noise-reduced audio signal is the first step in analyzing the residual noise. Spectral analysis can clearly reveal the distribution of noise in different frequency bands.

[0100] For example, in a quiet office environment, the audio signal after noise reduction may still contain low-frequency humming and high-frequency current noise from the operation of the air conditioner. Assume that through spectrum analysis, it is found that there is a low-frequency noise with an amplitude of about -15dB at 50Hz, and the high-frequency noise at 3kHz has an amplitude of -18dB. This distribution reflects the characteristics of the residual noise and provides a basis for subsequent processing.

[0101] In one possible implementation, when determining the residual noise intensity, the preset threshold can be set to -20dB. If -15dB at 50Hz exceeds this threshold, it indicates that the low-frequency noise is still significant and needs further optimization. It should be noted that it is a common method to adjust the tool to update the Q value parameter. The Q value affects the bandwidth of the filter. A higher Q value makes the filter more accurate but narrower, while a lower Q value covers a wider range but may affect the target signal. For example, adjusting the Q value from 1.0 to 1.5 and designing a narrowband filter for 50Hz noise can more concentratedly weaken the noise in this frequency band.

[0102] Specifically, the optimized time domain response data needs to be re-evaluated. Assume that after adjusting the Q value, the amplitude at 50Hz drops to -22dB, and the low-frequency noise intensity is weakened. It is understandable that if the -20dB noise reduction standard is still not met at this time, it is necessary to repeat the adjustment. For example, the Q value is further increased to 2.0 to observe whether the time domain waveform is smoother and the noise spikes are reduced. This iterative process ensures that the noise is gradually controlled.

[0103] Preferably, for high frequency bands such as -18 dB noise at 3 kHz, if it also exceeds the threshold, different strategies may be adopted.

[0104] In one embodiment, the Q value is reduced to 0.8 to form a wider filtering range, because high-frequency noise is often scattered. This differentiated adjustment can take into account the characteristics of different frequency bands and avoid a single parameter affecting the overall effect. For example, low frequencies are focused and suppressed with a high Q value, and high frequencies are smoothed with a low Q value. The combination of the two makes the audio cleaner.

[0105] In one embodiment, if a knocking sound suddenly occurs in an office environment, the spectrum shows a brief rise to -10dB at 1kHz, which exceeds the threshold by a large margin. At this time, the Q value can be dynamically adjusted to 1.2, while shortening the filter response time to quickly weaken the sudden noise.

[0106] It is understandable that this flexibility improves the adaptability of the solution to non-stationary noise and helps maintain the stability of the audio. From multiple perspectives, if the initial Q value is set to a fixed value such as 1.0, it may not be able to cope with noise with complex frequency distribution; if the amplitude change trend is combined in the iterative adjustment, such as 50Hz noise gradually decreasing from -15dB to -22dB, the Q value step size can be optimized accordingly. Exemplarily, the step size is set to 0.1, gradually approaching the optimal parameters. This refinement can more accurately match the noise characteristics.

[0107] Preferably, real-time spectrum monitoring is added to verify the effect immediately after each adjustment to ensure that the parameter configuration is gradually improved and ultimately meets the noise reduction standard.

[0108] The embodiment of the present invention also provides a noise reduction processing system for a Bluetooth audio device, the system comprising: a spectrum distribution data generation module, which obtains an audio signal from the surrounding environment of the Bluetooth audio device, performs a fast Fourier transform operation on the audio signal, and generates spectrum distribution data; a frequency feature set generation module, which determines whether there is a multi-source interference component exceeding a preset threshold value based on the intensity value of each frequency band in the spectrum distribution data, and generates a frequency feature set of the multi-source interference component if so; a dynamic fluctuation parameter determination module, which calculates the rate of change of the intensity of each frequency band in a time domain window for the frequency feature set, and determines the corresponding dynamic fluctuation parameter; a filter configuration adjustment module, which adjusts the filter configuration according to the corresponding relationship between the dynamic fluctuation parameter and the preset noise width; a time domain response parameter update module, which synchronously updates the time domain response parameter of the filter based on the amplitude change rate of the dynamic fluctuation parameter; and a dynamic threshold sequence update module, The upper and lower limits of the dynamic threshold sequence are calculated according to the time domain response parameters. When it is detected that the spectrum distribution data exceeds the upper limit of the dynamic threshold sequence, the threshold increment adjustment operation is triggered and the dynamic threshold sequence is updated; the noise reduction audio signal generation module performs gain compensation processing on the audio signal based on the updated dynamic threshold sequence to generate a noise reduction audio signal; the residual noise processing module detects the residual noise intensity of the noise reduction audio signal. If the residual noise intensity exceeds the set standard, the filter configuration adjustment and the time domain response parameter update operation are repeatedly performed until the noise reduction requirements are met.

[0109] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A noise reduction processing method for a Bluetooth audio device, characterized in that: The method comprises: The method comprises: obtaining an audio signal of the surrounding environment of the Bluetooth audio device, performing a fast Fourier transform operation on the audio signal, and generating spectrum distribution data; judging whether there is a multi-source interference component exceeding a preset threshold value according to the intensity value of each frequency band in the spectrum distribution data, and generating a frequency feature set of the multi-source interference component if so; calculating the rate of change of the intensity of each frequency band in the time domain window for the frequency feature set, and determining the corresponding dynamic fluctuation parameter; adjusting the filter configuration according to the corresponding relationship between the dynamic fluctuation parameter and the preset noise width; synchronously updating the time domain response parameter of the filter based on the amplitude change rate of the dynamic fluctuation parameter; calculating the upper and lower limits of the dynamic threshold sequence according to the time domain response parameter, and triggering a threshold increment adjustment operation and updating the dynamic threshold sequence when it is detected that the spectrum distribution data exceeds the upper limit of the dynamic threshold sequence; performing gain compensation processing on the audio signal based on the updated dynamic threshold sequence to generate a noise reduction audio signal; detecting the residual noise intensity of the noise reduction audio signal, and if the residual noise intensity exceeds the set standard, repeatedly performing the filter configuration adjustment and the time domain response parameter update operation until the noise reduction requirement is met.

2. The method according to claim 1, characterized in that: The step of acquiring an audio signal from the surrounding environment of the Bluetooth audio device and performing a fast Fourier transform operation on the audio signal to generate spectrum distribution data includes: Acquire the audio signal of the surrounding environment of the Bluetooth audio device, capture the original waveform through a sensor, and obtain initial audio data; Using filtering technology to remove noise from the initial audio data to obtain purified audio data; A fast Fourier transform is performed on the cleaned audio data to generate frequency spectrum distribution data.

3. The method according to claim 1, characterized in that The determining, based on the intensity value of each frequency band in the spectrum distribution data, whether there is a multi-source interference component exceeding a preset threshold, and if so, generating a frequency feature set of the multi-source interference component, comprises: The intensity value of each frequency band after division is obtained through the spectrum distribution data, and a preset threshold is used to determine whether there is an intensity value of the frequency band exceeding the threshold. If so, a preliminary identification result of multi-source interference is further obtained; Based on the preliminary identification result, fast Fourier transform is used to extract the frequency characteristics of the interference component to generate a corresponding frequency characteristic set; According to the frequency feature set, a preset threshold is used for screening to determine the distribution data of the multi-source interference; By using the distribution data of the multi-source interference, a time series analysis method is used to analyze the change trend of the spectrum distribution data, and the dynamic characteristics of the interference component are obtained; According to the dynamic characteristics, support vector machine algorithm is used for classification to obtain the category characteristics of the interference component; According to the category characteristics, the fluctuation range of the signal strength is extracted, and the frequency band division algorithm is used to divide the multi-source interference into frequency bands. Through the frequency band division, the boundary conditions of the frequency feature set are adjusted to obtain the optimized frequency feature set.

4. The method according to claim 1, characterized in that: The step of calculating the change rate of each frequency band intensity in the time domain window for the frequency feature set and determining the corresponding dynamic fluctuation parameter includes: Obtaining original frequency band intensity data from the frequency feature set, and using a fixed-length sliding window division method to generate time domain window sequence data; Based on the time domain window sequence data, using the first order difference method, the rate of change of the intensity of each frequency band is calculated; Determining whether the change rate exceeds a preset threshold range, and if so, adjusting the dynamic fluctuation range; Based on the fluctuation pattern within the dynamic fluctuation range, a K-means clustering algorithm is used to perform classification to obtain intensity distribution categories; Extracting the frequency band intensity fluctuation characteristics of each intensity distribution category from the time domain window sequence, and determining the boundary adjustment parameters of the frequency feature set; Adjusting the frequency feature set based on the boundary adjustment parameter, and analyzing the trend of the window sequence data based on the adjusted frequency feature set to obtain a period parameter; According to the periodic parameters, the mean and variance of the raw data of the frequency band intensity are calculated to determine the corresponding dynamic fluctuation parameters.

5. The method according to claim 1, characterized in that The adjusting the filter configuration according to the correspondence between the dynamic fluctuation parameter and the preset noise width includes: Acquire a dynamic fluctuation parameter and a noise width value, and search for an adjustment coefficient corresponding to the dynamic fluctuation parameter and the noise width value from a parameter mapping table; For a preset frequency band, generating an adjustment sequence corresponding to each frequency band according to the adjustment coefficient; If the adjustment sequence exceeds the preset range, a linear interpolation method is used to correct the adjustment sequence to obtain a corrected adjustment sequence; The initial configuration parameters of the filter are generated according to the modified adjustment sequence, and the configuration adjustment values ​​are calculated according to the matching degree between the initial configuration parameters and the dynamic fluctuation parameters. The initial configuration parameters are optimized by the configuration adjustment values ​​to obtain the optimized filter configuration.

6. The method according to claim 1, characterized in that The synchronously updating the time domain response configuration of the filter based on the amplitude change rate of the dynamic fluctuation parameter includes: Acquire the instantaneous value based on the change in the amplitude of the dynamic fluctuation parameter, determine the distribution characteristics of the change rate, and generate a preliminary adjustment strategy for the time domain response according to the distribution characteristics; Acquire a quantization sequence required for adjustment according to the preliminary adjustment strategy, use the quantization sequence to determine a deviation value of a response parameter, and establish an initial mapping table based on a mapping relationship between the deviation value and a time constant; The time constant is optimized through the initial mapping table, and the filter configuration parameters are updated using the optimized time constant to obtain the final time domain response configuration.

7. The method according to claim 1, characterized in that The step of calculating the upper and lower limits of the dynamic threshold sequence according to the time domain response parameter, and triggering a threshold increment adjustment operation and updating the dynamic threshold sequence when it is detected that the spectrum distribution data exceeds the upper limit of the dynamic threshold sequence, comprises: Acquire the time domain response configuration, and calculate the upper limit and lower limit of the dynamic threshold sequence to determine the detection range of the spectrum distribution data; If the spectrum distribution data exceeds the upper limit, a threshold increment adjustment is triggered to obtain an adjusted threshold increment; updating the dynamic threshold sequence using the adjusted threshold increment to obtain updated upper and lower limits; The adjustment is continuously triggered according to the updated upper and lower limits until the dynamic threshold sequence satisfying the stability condition is obtained.

8. A noise reduction processing system for a Bluetooth audio device, characterized in that: The system comprises: A spectrum distribution data generation module, which obtains the audio signal of the surrounding environment of the Bluetooth audio device, performs a fast Fourier transform operation on the audio signal, and generates spectrum distribution data; a frequency feature set generation module, which determines whether there is a multi-source interference component exceeding a preset threshold value based on the intensity value of each frequency band in the spectrum distribution data, and generates a frequency feature set of the multi-source interference component if so; a dynamic fluctuation parameter determination module, which calculates the rate of change of the intensity of each frequency band in the time domain window for the frequency feature set, and determines the corresponding dynamic fluctuation parameter; a filter configuration adjustment module, which adjusts the filter configuration according to the corresponding relationship between the dynamic fluctuation parameter and the preset noise width; a time domain response parameter update module, which synchronously updates the time domain response parameter of the filter based on the amplitude change rate of the dynamic fluctuation parameter; a dynamic threshold sequence update module, The upper and lower limits of the dynamic threshold sequence are calculated according to the time domain response parameters. When it is detected that the spectrum distribution data exceeds the upper limit of the dynamic threshold sequence, the threshold increment adjustment operation is triggered and the dynamic threshold sequence is updated; the noise reduction audio signal generation module performs gain compensation processing on the audio signal based on the updated dynamic threshold sequence to generate a noise reduction audio signal; the residual noise processing module detects the residual noise intensity of the noise reduction audio signal. If the residual noise intensity exceeds the set standard, the filter configuration adjustment and the time domain response parameter update operation are repeatedly performed until the noise reduction requirements are met.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the noise reduction processing method for a Bluetooth audio device according to any one of claims 1 to 7 is implemented.

Images