Motorcycle riding earphone different noise reduction algorithm combination noise reduction method

Abstract

The application relates to the technical field of earphone noise reduction processing, and discloses a motorcycle riding earphone different noise reduction algorithm combination noise reduction method, which comprises the following steps: obtaining an original audio signal and distributing the original audio signal to each processing module; generating a pretreatment signal, determining speech activity information, and calculating environmental noise energy information; adaptively adjusting high and low noise energy thresholds according to the environmental noise energy or auxiliary information; inputting the original signal into traditional and AI noise reduction modules in parallel to generate two-way noise reduction signals; generating an output signal through weighted fusion based on the environmental noise, speech information and thresholds; and performing equalization and compression processing on the output signal and driving a loudspeaker to output. The application introduces comprehensive judgment on environmental noise energy information, speech activity information and riding speed, the output signal of the traditional noise reduction module is set to zero in a high-speed non-speech environment, and AI noise reduction is combined, so that the sound quality distortion of the traditional noise reduction algorithm in a noise environment is effectively avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of earphone noise reduction processing, in particular to a motorcycle riding earphone different noise reduction algorithm combination noise reduction method. BACKGROUND

[0002] During motorcycle riding, the earphone worn by the rider is continuously exposed to a variable and high-intensity acoustic environment. Wind noise, engine roar, road friction sound, and surrounding vehicle noise constitute the main interference sources. These noises seriously affect the rider's reception of navigation instructions, communication, or enjoyment of music.

[0003] The existing noise reduction scheme mainly includes traditional noise reduction algorithms and artificial intelligence (AI) based noise reduction algorithms. Traditional noise reduction algorithms, such as techniques based on spectral subtraction or Wiener filtering, have the advantage of low computational overhead and obvious suppression effect on stable noise. In practical applications, traditional algorithms are often used to process scenes with relatively stable background noise, such as indoor communication or voice enhancement in a fixed environment.

[0004] However, in the process of motorcycle riding, the performance of traditional noise reduction algorithms significantly decreases when facing non-stationary, wideband wind noise generated by high-speed motorcycle riding. Noise estimation relies on the accuracy of voice activity detection, and once misjudgment occurs, it leads to excessive suppression of voice. Moreover, AI models require a large amount of training data and high generalization ability, and their performance is limited when facing unknown noise types. Therefore, the present application provides a motorcycle riding earphone different noise reduction algorithm combination noise reduction method to solve the deficiencies in the prior art. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a motorcycle riding earphone different noise reduction algorithm combination noise reduction method, which solves the problem of poor performance of traditional noise reduction methods and existing noise reduction schemes in dealing with complex and variable noise environments in motorcycle riding scenarios.

[0006] To achieve the above purpose, the present application is implemented through the following technical solutions:

[0007] The present application provides a motorcycle riding earphone different noise reduction algorithm combination noise reduction method, comprising the following steps:

[0008] S1, obtain the original input audio signal containing voice and environmental noise through a microphone, and copy and distribute the signal to multiple processing paths, including a weak noise reduction processing module, a noise energy detection module, a traditional noise reduction module, and an AI noise reduction module.

[0009] S2, a light noise reduction processing is performed on the audio signal received by the weak noise reduction processing module to generate a preprocessed signal. Based on the preprocessed signal, a high-precision voice activity detection algorithm is used to determine whether there is voice activity in the current time frame, so as to extract and determine the voice activity information. At the same time, based on the original input audio signal received by the noise energy detection module, when the voice activity information indicates no voice activity, the instantaneous noise energy of the current time frame is calculated, and by performing first-order low-pass smoothing filtering processing on the instantaneous noise energy, the stable environmental noise energy information is obtained. The calculation formula of the instantaneous noise energy is:

[0010]

[0011] In the formula, X(f, t) represents the spectral component of the original input audio signal at the frequency index f and the time frame index t,

[0012] S3, according to the change trend of the environmental noise energy information or external auxiliary information, the low noise energy threshold and the high noise energy threshold for distinguishing the riding speed state are adaptively determined and set, which improves the environmental adaptability of the system.

[0013] S4, the original input audio signal is input into the traditional noise reduction module and the AI noise reduction module for processing, respectively generating a first noise reduction signal and a second noise reduction signal. The traditional noise reduction module can use the optimal modified log spectrum amplitude (OMLSA) algorithm or its optimized variant for processing. The AI noise reduction module is processed by a deep learning noise reduction algorithm, when the voice activity information indicates no voice activity and the environmental noise energy information is higher than the dynamically adjusted high noise energy threshold, the first noise reduction signal output by the traditional noise reduction module is set to zero, so as to realize the adjustment of the traditional noise reduction effect in the specific high-speed voiceless scene.

[0014] ​​​​​​​​​​S5, intelligently select the first noise reduction signal or the second noise reduction signal as the final output signal according to the ambient noise energy information, the voice activity information, and the determined low noise energy threshold and high noise energy threshold, when the ambient noise energy information is not higher than the dynamically adjusted low noise energy threshold, select the second noise reduction signal as the output signal; when the ambient noise energy information is higher than the dynamically adjusted high noise energy threshold and the voice activity information indicates that there is voice activity, select the first noise reduction signal as the output signal; when the ambient noise energy information is higher than the dynamically adjusted high noise energy threshold and the voice activity information indicates that there is no voice activity, set the first noise reduction signal to zero as the output signal. When the ambient noise energy information is in the transition interval between the dynamically adjusted low noise energy threshold and the high noise energy threshold, the first noise reduction signal and the second noise reduction signal are weighted and fused to generate the output signal, wherein the weighting coefficient is dynamically adjusted according to the ambient noise energy information or the voice activity information.

[0015] S6, equalization processing and dynamic range compression processing are performed on the generated output signal to optimize the audio quality, and the optimized audio signal is driven to the loudspeaker for output, thereby completing the entire noise reduction process.

[0016] The second aspect of the application provides a motorcycle riding earphone different noise reduction algorithm combination noise reduction system, which is applied to the above method and includes:

[0017] A microphone is configured to acquire an original input audio signal including voice and ambient noise.

[0018] A weak noise reduction processing module is configured to perform weak noise reduction processing on the original input audio signal to generate a preprocessed signal.

[0019] A voice activity detection module is configured to determine voice activity information based on the preprocessed signal.

[0020] A noise energy detection module is configured to calculate instantaneous noise energy based on the original input audio signal and the voice activity information, and generate ambient noise energy information through smoothing processing.

[0021] A conventional noise reduction module is configured to perform conventional noise reduction processing on the original input audio signal to generate a first noise reduction signal.

[0022] An AI noise reduction module is configured to perform AI noise reduction processing on the original input audio signal to generate a second noise reduction signal.

[0023] A threshold adjustment module is configured to adaptively adjust and establish a low noise energy threshold and a high noise energy threshold for judging the riding speed according to the change trend of the ambient noise energy information or external auxiliary information.

[0024] a selection and fusion module configured to generate an output signal by selecting or weighting fusing the first noise reduction signal and the second noise reduction signal according to the ambient noise energy information, the voice activity information, and the low noise energy threshold and the high noise energy threshold.

[0025] a post-processing module configured to perform equalization and dynamic range compression processing on the output signal to optimize audio quality.

[0026] a speaker configured to drive the optimized audio signal output.

[0027] The present application provides a motorcycle riding earphone different noise reduction algorithm combination noise reduction method. It has the following beneficial effects:

[0028] 1. The present application realizes the adaptive selection and weighting fusion of traditional noise reduction algorithm and AI noise reduction algorithm by introducing the comprehensive judgment of ambient noise energy information, voice activity information and riding speed. By setting the output signal of the traditional noise reduction module to zero in a high-speed non-speech environment, combined with AI noise reduction, the audio distortion caused by traditional noise reduction algorithm in noisy environment is effectively avoided, and the wind noise and other high-speed environmental noise are maximally suppressed, so that more accurate and efficient noise reduction experience can be provided under different riding speeds and noise intensities.

[0029] 2. The present application generates a preprocessed signal through a weak noise reduction processing module, and judges voice activity information based on the preprocessed signal by a voice activity detection module. When the noise energy detection module calculates the instantaneous noise energy, the voice activity information is used to ensure that the instantaneous noise energy is calculated only when there is no voice activity, and the ambient noise energy information is generated by first-order low-pass smoothing filtering processing, which can effectively avoid the interference of voice component on noise estimation, thereby improving the accuracy of ambient noise energy information.

[0030] 3. The present application dynamically adjusts the low noise energy threshold and the high noise energy threshold, and intelligently selects traditional noise reduction, AI noise reduction or fusion of the two on this basis. In low-speed or voice-dominated scenarios, AI noise reduction can be selected to obtain better audio quality and voice fidelity; in high-speed or pure noise scenarios, by adjusting the output of traditional noise reduction, unnecessary AI noise reduction calculation overhead can be avoided while ensuring the noise reduction effect, thereby effectively balancing the noise reduction performance and system resource consumption in different scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The motorcycle riding earphone different noise reduction algorithm combination noise reduction system architecture of the present application;

[0032] Figure 2 The method flowchart of the present application;

[0033] Figure 3 Figure 1 is a schematic diagram of an algorithmic framework of the present application.

[0034] wherein 101, microphone; 102, weak noise reduction processing module; 103, noise energy detection module; 104, traditional noise reduction module; 105, AI noise reduction module; 106, voice activity detection module; 107, threshold adjustment module; 108, selection and fusion module; 109, post-processing module; 110, speaker. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0036] Referring to the drawings Figure 1 , Figure 1 Figure 1 is a schematic diagram of a motorcycle riding earphone different noise reduction algorithm combination noise reduction system architecture according to an embodiment of the present application. The present application provides a motorcycle riding earphone different noise reduction algorithm combination noise reduction system, which comprises a microphone 101, a weak noise reduction processing module 102, a noise energy detection module 103, a traditional noise reduction module 104, an AI noise reduction module 105, a voice activity detection module 106, a threshold adjustment module 107, a selection and fusion module 108, a post-processing module 109, and a speaker 110.

[0037] The original input audio signal is collected by the microphone 101, which contains voice and environmental noise. The original input audio signal collected by the microphone 101 is copied and distributed to the weak noise reduction processing module 102, the noise energy detection module 103, the traditional noise reduction module 104, and the AI noise reduction module 105. This parallel distribution mode ensures that different processing paths can process the original signal at the same time, improving the system processing efficiency.

[0038] After receiving the original input audio signal, the weak noise reduction processing module 102 performs weak noise reduction processing on it to generate a preprocessed signal. The preprocessed signal is sent to the voice activity detection module 106. The voice activity detection module 106 determines whether there is voice activity in the current time frame based on the preprocessed signal and outputs voice activity information. This voice activity information is sent as input to the noise energy detection module 103 and the selection and fusion module 108.

[0039] The noise energy detection module 103 receives the raw input audio signal and the voice activity information. Based on these inputs, the noise energy detection module 103 calculates the instantaneous noise energy and generates the ambient noise energy information through smoothing processing. This ambient noise energy information is transmitted to the threshold adjustment module 107 and the selection and fusion module 108.

[0040] The conventional noise reduction module 104 and the AI noise reduction module 105 receive the raw input audio signal in parallel. The conventional noise reduction module 104 performs conventional noise reduction processing on the raw input audio signal to generate a first noise reduction signal. The AI noise reduction module 105 performs AI noise reduction processing on the raw input audio signal to generate a second noise reduction signal. Both the first noise reduction signal and the second noise reduction signal are sent to the selection and fusion module 108.

[0041] The threshold adjustment module 107 receives the ambient noise energy information and can optionally receive external auxiliary information (such as speed data from a motorcycle speed sensor or a positioning system). Based on this information, the threshold adjustment module 107 adaptively adjusts and establishes the low noise energy threshold and the high noise energy threshold for judging the riding speed state. These dynamically adjusted thresholds are transmitted to the selection and fusion module 108.

[0042] The selection and fusion module 108 is the core decision-making unit of the system. This module receives the ambient noise energy information, the voice activity information, the low noise energy threshold, the high noise energy threshold, the first noise reduction signal, and the second noise reduction signal. According to these inputs, the selection and fusion module 108 generates the final output signal through selection or weighted fusion.

[0043] The final output signal is sent to the post-processing module 109. The post-processing module 109 performs equalization processing and dynamic range compression processing on the output signal to optimize the audio quality. The optimized audio signal is finally driven by the loudspeaker 110 for output, for the user to listen to. This system architecture ensures the integrity and adaptability of signal processing.

[0044] Referring to the accompanying drawings Figure 2 - the accompanying drawings Figure 3 The present application provides a motorcycle riding earphone different noise reduction algorithm combination noise reduction method, comprising the following steps:

[0045] S1, the raw input audio signal including voice and ambient noise is obtained through the microphone 101 and distributed to the weak noise reduction processing module 102, the noise energy detection module 103, the conventional noise reduction module 104, and the AI noise reduction module 105;

[0046] S2, weakly denoising the audio signal of the weak noise reduction processing module 102 to generate a preprocessed signal, determining voice activity information, and calculating the instantaneous noise energy based on the original input audio signal obtained by the noise energy detection module 103 and the voice activity information, and generating the ambient noise energy information through smoothing processing;

[0047] S3, according to the generated ambient noise energy information change trend or external auxiliary information, adaptively adjusting and establishing the low noise energy threshold and the high noise energy threshold for judging the riding speed;

[0048] S4, inputting the obtained original input audio signal into the traditional denoising module 104 and the AI denoising module 105 (Transformer architecture AI denoising) respectively for processing, and generating a first denoising signal and a second denoising signal respectively;

[0049] S5, according to the ambient noise energy information, the voice activity information, and the adjusted and established low noise energy threshold and high noise energy threshold, generating an output signal through weighted fusion;

[0050] S6, equalizing and dynamically compressing the generated output signal to optimize the audio quality, and driving the loudspeaker 110 to output.

[0051] For step S1, in the original signal acquisition and distribution step, the whole processing flow starts from the collection of the environmental acoustic signal, which is collected by one or more microphones 101 configured on the motorcycle riding earphone in real time, so as to obtain the original input audio signal containing voice and environmental noise. The original input audio signal is a time domain signal. In order to facilitate subsequent noise reduction processing, the time domain signal is converted into a time-frequency domain representation. This conversion process can be realized by short-time Fourier transform (STFT), which divides the time domain signal into multiple overlapping time frames and performs Fourier transform on each time frame to obtain the frequency spectrum component of the signal.

[0052] The time-frequency domain representation of the original input audio signal is input to the weak noise reduction processing module 102, the noise energy detection module 103, the traditional denoising module 104, and the AI denoising module 105 (Transformer architecture AI denoising) at the same time. This parallel distribution design ensures that each processing module can process the same basic signal at the same time, providing a technical basis for subsequent noise reduction strategy combination. The representation of the original input audio signal in the time-frequency domain can be represented as:

[0053] ;

[0054] In the formula, represents the original input audio signal in the time frame and the time index time-domain sample value at the time frame, denotes a short-time Fourier transform, denotes a spectral component of the original input audio signal at the frequency index and the time frame index .

[0055] For step S2, in the generation of the ambient noise and speech activity information, the process first performs a light noise reduction algorithm on the received original input audio signal in the weak noise reduction processing module 102, thereby generating a pre-processed signal. This light noise reduction aims to preliminarily reduce the noise so that the subsequent speech activity detection module 106 can more accurately identify the speech.

[0056] The pre-processed signal is transmitted to the speech activity detection module 106. The speech activity detection module 106 uses a voice activity detection (VAD) algorithm to determine whether there is speech activity in the current time frame based on the pre-processed signal. The output of the VAD algorithm is the speech activity information , which is usually represented in binary form: when , it indicates that there is speech in the current time frame; when 0, it indicates that there is no speech in the current time frame.

[0057] The noise energy detection module 103 receives the time-frequency domain representation of the original input audio signal and the speech activity information output by the speech activity detection module 106. When the speech activity information indicates no speech activity (i.e. ), the noise energy detection module 103 calculates the instantaneous noise energy of the current time frame. The calculation of the instantaneous noise energy uses the weighted instantaneous energy calculation formula:

[0058] ;

[0059] In the formula, S denotes a spectral component of the original input audio signal at the frequency index and the time frame index , denotes the frequency index, denotes the speech activity information in the time frame , denotes the presence of speech, denotes the absence of speech.

[0060] In order to obtain more stable and reliable ambient noise energy information, the calculated instantaneous noise energy is subjected to a first-order low-pass smoothing filter processing. The ambient noise energy information after smoothing filtering may be expressed as:

[0061] ;

[0062] wherein, represents the ambient noise energy information of the current time frame , represents the ambient noise energy information of the previous time frame , is a smoothing factor between 0 and 1, used to control the degree of smoothing. A larger value indicates stronger smoothing, resulting in a more gradual change in noise energy information. Through such smoothing, the influence of instantaneous noise fluctuations on system decision-making can be effectively reduced, providing a more stable ambient noise energy reference.

[0063] For step S3, in the adaptive adjustment of dynamic threshold step, the threshold adjustment module 107 adaptively determines and sets the low noise energy threshold and high noise energy threshold for distinguishing the riding speed state according to the trend of change of ambient noise energy information or the received external auxiliary information.

[0064] Specifically, the threshold adjustment module 107 calculates a noise reference adjustment amount according to the average value or rate of change of ambient noise energy information over a predetermined period of time. For example, when the ambient noise energy information shows a consistent upward trend over a period of time, it may indicate that the riding speed is increasing, and the noise reference adjustment amount will be a positive value.

[0065] In addition, the threshold adjustment module 107 can also receive external auxiliary information such as speed data provided by a motorcycle speed sensor or positioning system . These external speed data can directly reflect the current riding speed, providing a more direct basis for threshold adjustment.

[0066] Based on the preset reference low noise energy threshold and reference high noise energy threshold , combined with the above calculated noise reference adjustment amount or external auxiliary information , the threshold adjustment module 107 adaptively adjusts the low noise energy threshold and high noise energy threshold . Taking the noise reference adjustment amount as an example, the dynamically adjusted threshold can be expressed as:

[0067] ;

[0068] wherein, denotes the dynamically adjusted threshold (which can be or ), denotes the preset reference threshold (which can be or ), is an adjustment function based on the noise reference adjustment amount . The adjustment function can be nonlinear, for example, when reaches a certain preset value, the amplitude of the threshold adjustment will increase to adapt to more significant speed changes. In this way, the low noise energy threshold and the high noise energy threshold can reflect the current cycling environment in real time, providing dynamic and accurate judgment basis for subsequent noise reduction strategy selection.

[0069] For step S4, in the parallel processing step of traditional noise reduction and AI noise reduction, the time-frequency domain representation of the original input audio signal is simultaneously input to the traditional noise reduction module 104 and the AI noise reduction module 105 (Transformer architecture AI noise reduction) for parallel processing, respectively generating a first noise reduction signal and a second noise reduction signal .

[0070] The traditional noise reduction module 104 processes the received original input audio signal to generate a first noise reduction signal. This module uses the optimal modified log spectrum amplitude algorithm for noise reduction. The OMLSA algorithm modifies the log spectrum amplitude of the noisy signal based on noise estimation and speech presence probability to suppress noise. The core of the OMLSA algorithm is to estimate the gain function , which adjusts the amplitude spectrum of the original signal. The modified log spectrum amplitude can be represented as:

[0071] ;

[0072] wherein, denotes the amplitude spectrum of the first noise reduction signal in the time-frequency domain after traditional noise reduction, denotes the amplitude spectrum of the original input audio signal in the time-frequency domain, is the gain function calculated by the OMLSA algorithm, whose value is between 0 and 1, used to attenuate noise. In specific implementation, the calculation of the gain function will consider the estimation of the noise spectrum and the speech presence probability to suppress noise while preserving the speech components as much as possible.

[0073] It is worth noting that under certain conditions, i.e., when the speech activity information indicates no speech activity (i.e. ) and environmental noise energy information above the dynamically adjusted high noise energy threshold In this case, the traditional noise reduction module 104 sets the output first noise reduction signal to zero. This processing aims to effectively avoid the introduction of musical noise or residual noise by the traditional noise reduction algorithm in the extreme noise and no-speech scenario, ensuring the purity of the signal output.

[0074] The AI noise reduction module 105 (Transformer architecture AI noise reduction) simultaneously receives the time-frequency domain representation of the original input audio signal , processes it through a deep learning noise reduction algorithm, and generates a second noise reduction signal . The deep learning model learns the mapping relationship from noisy speech to clean speech through training, and can handle complex non-stationary noise. The generation process of the second noise reduction signal can be summarized as:

[0075] ;

[0076] wherein, represents a deep neural network model, the input of which is the spectral component of the original input audio signal, and the output is the spectral component of the second noise reduction signal after deep learning noise reduction processing. This deep learning model is usually trained on a large set of noisy speech and clean speech data through supervised learning to learn the characteristics of noise and effectively suppress it, while preserving the clarity and naturalness of speech to the greatest extent.

[0077] The first noise reduction signal and the second noise reduction signal output in parallel by the traditional noise reduction module 104 and the AI noise reduction module 105 (Transformer architecture AI noise reduction) will be the input of the subsequent selection and fusion module 108 for intelligent decision of the final output signal.

[0078] For step S5, in the intelligent selection and weighted fusion decision step, the selection and fusion module 108 receives the environmental noise energy information from the noise energy detection module 103, the speech activity information from the speech activity detection module 106, the low noise energy threshold and the high noise energy threshold from the threshold adjustment module 107, and the first noise reduction signal from the traditional noise reduction module 104 and the second noise reduction signal from the AI noise reduction module 105 (Transformer architecture AI noise reduction). According to these input information, this module intelligently decides the final output signal.

[0079] The selection and fusion module 108 executes the following strategies according to the comparison relationship between the ambient noise energy information and the dynamically adjusted low noise energy threshold and high noise energy threshold, combined with the voice activity information:

[0080] When the ambient noise energy information is not higher than the dynamically adjusted low noise energy threshold, the system determines that it is currently in a low noise environment. In this case, the selection and fusion module 108 directly selects the second noise reduction signal as the output signal. The AI noise reduction algorithm can usually provide more natural and high-fidelity sound quality in a low noise environment.

[0081] When the ambient noise energy information is higher than the dynamically adjusted high noise energy threshold and the voice activity information indicates that there is voice activity, the system determines that it is currently in a high noise and voice environment. In this case, the selection and fusion module 108 directly selects the first noise reduction signal as the output signal. The robustness and voice protection ability of the traditional noise reduction algorithm are embodied when there is strong noise and voice.

[0082] When the ambient noise energy information is higher than the dynamically adjusted high noise energy threshold and the voice activity information indicates that there is no voice activity, the system determines that it is currently in a high noise and voice-free environment, and the first noise reduction signal has been set to zero by the traditional noise reduction module 104. The selection and fusion module 108 at this time will take the zeroed first noise reduction signal as the output signal, realizing the maximum suppression of noise.

[0083] When the ambient noise energy information is between the dynamically adjusted low noise energy threshold and the high noise energy threshold , the selection and fusion module 108 performs weighted fusion on the first noise reduction signal and the second noise reduction signal to generate the output signal . The weighted fusion can be expressed as:

[0084] ;

[0085] In the formula, represents the output spectrum of the th frequency unit, the th frame; the spectrum estimation obtained based on the low noise environment training; the spectrum estimation obtained based on the high noise environment training; the noise energy or ambient information of the current frame , the weight coefficient adaptively calculated.

[0086] For step S6, in the post-processing and signal output step, the selection and fusion module 108 generates the final output signal The signal is sent to a post-processing module 109. The post-processing module 109 performs further optimization on the signal to ensure that the final output audio quality meets the listening requirements.

[0087] The post-processing module 109 first performs equalization processing on the output signal. Equalization processing aims to adjust the spectral characteristics of the signal to compensate for the frequency response imbalance caused by the microphone, speaker, and environmental acoustic characteristics. For example, a multi-section equalizer can be used to adjust the gain of different frequency bands, so that the final audio output achieves a more flat frequency response curve across the entire frequency band, making the sound more natural.

[0088] The post-processing module 109 performs dynamic range compression processing on the equalized signal. Dynamic range compression is used to reduce the difference between the highest and lowest parts of the signal, so that the quiet voice part is easier to hear without distortion, and the discomfort caused by excessive loudness is avoided. Dynamic range compression is achieved by setting parameters such as compression threshold and compression ratio.

[0089] After equalization processing and dynamic range compression processing, the optimized audio signal is sent to the speaker 110. The speaker 110 converts the received electrical signal into sound waves, thus playing the final noise-reduced audio signal to the user. The entire processing process is completed, providing the user with a noise-reduced listening experience.

Claims

1. A motorcycle riding earphone noise reduction method combining different noise reduction algorithms, characterized in that, The method comprises the following steps: S1, obtaining an original input audio signal including voice and environmental noise through a microphone, and distributing to a weak noise reduction processing module, a noise energy detection module, a traditional noise reduction module, and an AI noise reduction module; S2, performing weak noise reduction processing on the audio signal of the weak noise reduction processing module to generate a preprocessed signal, determining voice activity information, and calculating instantaneous noise energy based on the original input audio signal obtained by the noise energy detection module and the voice activity information, and generating environmental noise energy information through smoothing processing; S3, adaptively adjusting and establishing a low noise energy threshold and a high noise energy threshold for judging the riding speed according to the generated environmental noise energy information change trend or external auxiliary information; S4, inputting the obtained original input audio signal to the traditional noise reduction module and the AI noise reduction module respectively for processing, and generating a first noise reduction signal and a second noise reduction signal respectively; S5, generating an output signal through weighted fusion according to the environmental noise energy information, the voice activity information, and the adjusted and established low noise energy threshold and high noise energy threshold; S6, performing equalization and dynamic range compression processing to optimize the audio quality, and driving the loudspeaker to output.

2. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 1, characterized in that, In step S1, the original input audio signal including voice and environmental noise obtained by the microphone comprises the following steps: Real-time acquisition of acoustic signals around the motorcycle riding earphone through a microphone to obtain the original input audio signal; Convert the original input audio signal into a time-frequency domain representation through short-time Fourier transform; Input the original input audio signal in the time-frequency domain representation to the weak noise reduction processing module, the noise energy detection module, the traditional noise reduction module, and the AI noise reduction module.

3. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 1, characterized in that, In step S2, the environmental noise energy information generated through smoothing processing comprises the following steps: Perform a lightweight noise reduction algorithm on the original input audio signal of the weak noise reduction processing module to generate the preprocessed signal; Based on the preprocessed signal, use a voice activity detection algorithm to determine whether there is voice activity in the current time frame, and further generate the voice activity information; Based on the original input audio signal received by the noise energy detection module, when the voice activity information indicates no voice activity, calculate the instantaneous noise energy of the current time frame, and perform first-order low-pass smoothing filter processing to generate the environmental noise energy information.

4. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 3, characterized in that, The instantaneous noise energy The formula for calculating the instantaneous noise energy is: ； wherein denotes a spectral component of the original input audio signal at frequency index and time frame index denotes a spectral component of the original input audio signal at frequency index denotes a frequency index, denotes a speech activity information at time frame denotes speech present, denotes speech present, denotes speech absent.

5. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 1, characterized in that, In step S3, the adaptive adjustment and establishment of the low noise energy threshold and the high noise energy threshold for judging the riding speed comprises the following steps: Calculate the noise reference adjustment amount according to the average value or the change rate of the environmental noise energy information within a preset time period; Receive speed data from a motorcycle speed sensor or a positioning system as the external auxiliary information; Based on the preset reference low noise energy threshold and the reference high noise energy threshold, and combined with the noise reference adjustment amount or the external auxiliary information, adaptively adjust the low noise energy threshold and the high noise energy threshold.

6. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 1, characterized in that, In step S4, the generation of the first noise reduction signal and the second noise reduction signal comprises the following steps: Process the original input audio signal received by the traditional noise reduction module through the optimal modified log spectrum amplitude algorithm to generate the first noise reduction signal; The original input audio signal received by the AI noise reduction module is processed by a deep learning noise reduction algorithm to generate the second noise reduction signal.

7. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 6, characterized in that, When the voice activity information indicates no voice activity and the environmental noise energy information is higher than the dynamically adjusted high noise energy threshold, the traditional noise reduction module sets the output first noise reduction signal to zero.

8. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 1, characterized in that, In step S5, when the environmental noise energy information is not higher than the dynamically adjusted low noise energy threshold, the second noise reduction signal is selected as the output signal; when the environmental noise energy information is higher than the dynamically adjusted high noise energy threshold and the voice activity information indicates voice activity, the first noise reduction signal is selected as the output signal.

9. The motorcycle riding earphone different noise reduction algorithm combination noise reduction method according to claim 1, characterized in that, In step S5, the output signal generated by the weighted fusion includes the following steps: When the environmental noise energy information is between the dynamically adjusted low noise energy threshold and the high noise energy threshold, the first noise reduction signal and the second noise reduction signal are weighted and fused to generate the output signal.

10. The motorcycle riding earphone different noise reduction algorithm combination noise reduction system applied to the motorcycle riding earphone different noise reduction algorithm combination noise reduction method of any one of claims 1-9, characterized in that, Comprise: A microphone for acquiring an original input audio signal including voice and environmental noise; A weak noise reduction processing module for weak noise reduction processing of the original input audio signal to generate a preprocessed signal; A voice activity detection module for determining voice activity information based on the preprocessed signal; A noise energy detection module for calculating instantaneous noise energy based on the original input audio signal and the voice activity information, and generating environmental noise energy information through smoothing processing; A traditional noise reduction module for traditional noise reduction processing of the original input audio signal to generate a first noise reduction signal; An AI noise reduction module for AI noise reduction processing of the original input audio signal to generate a second noise reduction signal; A threshold adjustment module for adaptively adjusting and establishing a low noise energy threshold and a high noise energy threshold for judging the riding speed according to the change trend of the environmental noise energy information or external auxiliary information; A selection and fusion module for selecting or weighting and fusing the first noise reduction signal and the second noise reduction signal to output a signal according to the environmental noise energy information, the voice activity information, and the low noise energy threshold and the high noise energy threshold; A post-processing module for equalization and dynamic range compression processing of the output signal to optimize the audio signal quality; A loudspeaker for driving the optimized audio signal output.

Images