A multi-microphone intelligent sports earphone noise reduction system and a noise reduction method thereof

Abstract

The application discloses a multi-microphone intelligent sports earphone noise reduction system, which comprises an audio acquisition module, a preprocessing module, a filtering unit, a sound-to-color mapping module, a wind noise suppression module and a noise reduction unit. The audio acquisition module is provided with at least three microphones. The filtering unit is used for generating a frequency band mask MaskFU1(f, t) based on a scene recognition result. The filtering unit filters non-target sound in a non-talking scene. The filtering unit further comprises frequency mapping as a hue H, energy mapping as a brightness L and voice existence probability or wind noise probability mapping as an opacity alpha. The wind noise suppression module is used for calculating a wind noise probability W(t) and dynamically adjusting a high-frequency attenuation factor according to color features output by the sound-to-color mapping module and the motion vibration signal. The noise reduction unit is used for combining the frequency band mask MaskFU1(f, t), the color spectrogram G C(f, t) and the wind noise probability W(t), and filtering an input signal through a frequency variable gain function Hfinal(f, t).

Description

Technical Field

[0001] This invention relates to the field of wireless headphones, and in particular to a multi-microphone intelligent sports headphone noise reduction system. Background Technology

[0002] With the development of wireless headphones, especially Bluetooth headphones, which can also include over-ear headphones, this system is mainly aimed at wireless Bluetooth headphones (including in-ear, semi-in-ear, clip-on, bone conduction headphones and other small headphones).

[0003] During the use of this type of small earphone, it is often found that there is a lot of noise, such as wind noise and external noise, whether it is a call or just listening to music, which seriously affects the comfort of use. The existing noise cancellation method is generally to use the built-in system for noise cancellation.

[0004] For example, Chinese patent CN202511022615.9 inputs the real and imaginary parts of the complex spectrum matrix of the noise audio collected by the microphone into a sliding converter network to obtain the predicted real and predicted imaginary parts; based on the complex spectrum composed of the predicted real and predicted imaginary parts, a ratio mask between the complex spectrum and the complex spectrum matrix is ​​calculated; the complex spectrum matrix is ​​enhanced based on the ratio mask to obtain the target complex spectrum; and an inverse short-time Fourier transform is performed on the target complex spectrum to obtain the noise-suppressed frequency.

[0005] The predicted output of the complex spectrum matrix is ​​calculated by the trained sliding converter network. Combined with the inference path of ratio mask fusion, the spectrum reconstruction of noisy audio is realized, avoiding noise cutting based directly on the audio energy threshold and improving the clarity of the denoised audio.

[0006] However, the actual use of a trained system and the implementation of spectrum conversion place high demands on the built-in chip, and the conversion noise reduction has a certain delay. Therefore, how to achieve accurate noise reduction and improve the accuracy of noise reduction without increasing chip cost, thereby improving the noise reduction effect, is the key to system design. Summary of the Invention

[0007] The main objective of this invention is to propose a multi-microphone intelligent sports headphone noise reduction system. The aim is to embed this system into a chip, thereby reducing the chip's data processing load (existing practices generally use high-priced processing chips to increase the noise reduction rate), thus improving noise reduction accuracy. Furthermore, this operation changes the existing single spectrum conversion method and is a better reference model in the field of acoustics. It can improve the accuracy of sound signal output or input and avoid blind spots in sound conversion (for example, it can achieve more accurate detection in acoustic testing).

[0008] To achieve the above objectives, this invention proposes a multi-microphone intelligent sports headphone noise reduction system, comprising: An audio acquisition module is provided with at least three microphones, which are used to acquire sound in the ear canal, ambient sound, and motion vibration signals, respectively. The preprocessing module is used to synchronize and calibrate the output signal of the audio acquisition module; A filtering unit is used to generate a frequency band mask MaskFU1(f,t) based on the scene recognition results. The filtering unit filters out non-target sounds in non-call scenarios, and the scene recognition results include call activation status, media playback status, and motion posture. The sound-to-color mapping module is used to convert time-domain audio signals into time-frequency domain color spectrograms Gc(f,t). This also includes mapping frequency to hue H, energy to brightness L, and mapping the probability of speech presence or wind noise to opacity α. A wind noise suppression module is used to calculate the wind noise probability W(t) based on the color features output by the sound-to-color mapping module and the motion vibration signal, and to dynamically adjust the high-frequency attenuation factor. A noise reduction unit is used to filter the input signal by combining the frequency band mask MaskFU1(f,t), the color spectrogram Gc(f,t) and the wind noise probability W(t) through the frequency-varying gain function Hfinal(f,t); An audio reconstruction module is used to convert the filtered signal into an analog audio output.

[0009] in, MaskFU1(f,t) is a technical term in the field of signal processing (especially audio noise reduction and speech enhancement), which stands for "frequency-varying mask of the filter unit".

[0010] (f,t) represents two independent variables: f(Frequency): Frequency; indicates that the mask operates in the frequency domain and has different processing strategies for different frequency bands.

[0011] t(Time): Time; indicating that the mask is time-varying and can be dynamically adjusted over time to adapt to real-time changes in the sound environment.

[0012] A noise reduction method for multi-microphone smart sports headphones, based on the aforementioned multi-microphone smart sports headphone noise reduction system, includes the following steps: S1 acquires multiple audio and motion signals and performs time-frequency conversion; S2, based on the current operating scenario of the device, execute the first filtering strategy and generate a frequency band mask to isolate non-call sounds; S3 maps the time-frequency signal into a color spectrogram that includes hue, brightness, and opacity; S4, analyze the blue / cyan band energy and sensor signal in the color spectrogram to estimate the wind noise probability; S5. Based on the physiological perception weights and wind noise probability corresponding to the color bands, the final frequency response mask is synthesized. S6, the original signal is weighted and reconstructed using the final frequency response mask.

[0013] 1. A three-dimensional noise reduction architecture of "filtering unit (FU1) + sound-to-color mapping + wind noise robustness" was constructed by using a multi-microphone audio acquisition module, a filtering unit, a sound-to-color mapping module, a wind noise suppression module, a noise reduction unit, and an audio reconstruction module. 2. The sound output or input is clearly defined by the filter unit (FU1), which solves the logical conflicts of smart headphones in complex switching scenarios (call / music / standby); especially the problems of circuit breakers, audio dropouts, or audio mixing. 3. Introducing "sound-to-color mapping logic" transforms abstract spectrum data into intuitive "color bands," which is beneficial for debugging; This can be understood as follows: abstract wavebands cannot accurately distinguish some abnormal sounds, but color capture can accurately acquire data, thereby improving recognition. In addition, this design can be transferred to a display screen, lighting system or mobile phone screen, thereby achieving a dual experience of sound and vision. 4. Through dual purification of input and output, processing of input microphone acquisition and output reconstruction modules ensures the clarity of uplink calls and the immersive experience of downlink playback; the output can also be adjusted according to external audio to adjust the low, mid and high frequency embellishments (i.e., increase the emphasis of a predetermined channel), thereby improving the output during use, i.e., improving auditory sensitivity. Of course, this part can be adjusted according to personal needs, such as sound testing, to improve the auditory quality or call quality during exercise. Attached Figure Description

[0014] Figure 1 This is the noise reduction logic diagram; Figure 2 A schematic diagram of a color spectrogram of a time-domain audio signal converted to the time-frequency domain; Figure 3 This is a schematic diagram of the color spectrum in the time-frequency domain after passing through a filtering unit. Detailed Implementation

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, top, bottom, inside, outside, vertical, horizontal, longitudinal, counterclockwise, clockwise, circumferential, radial, axial, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0017] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0018] like Figures 1 to 3 As shown, a multi-microphone smart sports headphone noise cancellation system includes: An audio acquisition module is provided with at least three microphones, which are used to acquire sound in the ear canal, ambient sound, and motion vibration signals, respectively. The preprocessing module is used to synchronize and calibrate the output signal of the audio acquisition module; A filtering unit is used to generate a frequency band mask MaskFU1(f,t) based on the scene recognition results. The filtering unit filters out non-target sounds in non-call scenarios, and the scene recognition results include call activation status, media playback status, and motion posture. The sound-to-color mapping module is used to convert time-domain audio signals into time-frequency domain color spectrograms Gc(f,t). This also includes mapping frequency to hue H, energy to brightness L, and mapping the probability of speech presence or wind noise to opacity α. A wind noise suppression module is used to calculate the wind noise probability W(t) based on the color features output by the sound-to-color mapping module and the motion vibration signal, and to dynamically adjust the high-frequency attenuation factor. A noise reduction unit is used to filter the input signal by combining the frequency band mask MaskFU1(f,t), the color spectrogram Gc(f,t) and the wind noise probability W(t) through the frequency-varying gain function Hfinal(f,t); An audio reconstruction module is used to convert the filtered signal into an analog audio output.

[0019] in, MaskFU1(f,t) is a technical term in the field of signal processing (especially audio noise reduction and speech enhancement), which stands for "frequency-varying mask of the filter unit".

[0020] (f,t) represents two independent variables: f(Frequency): Frequency; indicates that the mask operates in the frequency domain and has different processing strategies for different frequency bands.

[0021] t(Time): Time; indicating that the mask is time-varying and can be dynamically adjusted over time to adapt to real-time changes in the sound environment.

[0022] 1. A three-dimensional noise reduction architecture of "filtering unit (FU1) + sound-to-color mapping + wind noise robustness" was constructed by using a multi-microphone audio acquisition module, a filtering unit, a sound-to-color mapping module, a wind noise suppression module, a noise reduction unit, and an audio reconstruction module. 2. The sound output or input is clearly defined by the filter unit (FU1), which solves the logical conflicts of smart headphones in complex switching scenarios (call / music / standby); especially the problems of circuit breakers, audio dropouts, or audio mixing. 3. Introducing "sound-to-color mapping logic" transforms abstract spectrum data into intuitive "color bands," which is beneficial for debugging; This can be understood as follows: abstract wavebands cannot accurately distinguish some abnormal sounds, but color capture can accurately acquire data, thereby improving recognition. In addition, this design can be transferred to a display screen, lighting system or mobile phone screen, thereby achieving a dual experience of sound and vision.

[0023] 4. Through dual purification of input and output, processing of input microphone acquisition and output reconstruction modules ensures the clarity of uplink calls and the immersive experience of downlink playback; the output can also be adjusted according to external audio to adjust the low, mid and high frequency embellishments (i.e., increase the emphasis of a predetermined channel), thereby improving the output during use, i.e., improving auditory sensitivity. Of course, this part can be adjusted according to personal needs, such as sound testing, to improve the auditory quality or call quality during exercise.

[0024] Specifically, the audio acquisition module includes: Feedback microphone (M0), which is located inside the ear canal and mainly covers the mid-low frequency band of 300Hz-3500Hz; A feedforward microphone (M1) is located on the outside of the earphone and mainly covers the high frequency band of 2000Hz-10000Hz. A voice microphone (M2) is located on the side of the ear stem near the mouth, and mainly covers the human voice sensitive frequency band of 500Hz-4000Hz; A differential pressure or acceleration sensor (M3) is used to collect airflow pressure difference or housing vibration signals as an auxiliary feature for wind noise detection.

[0025] Feedback microphone (M0): Located inside the ear canal, it utilizes its sealed cavity characteristics to primarily pick up the mid-to-low frequency human voice bone conduction / air conduction residuals in the range of 300Hz to 3500Hz, serving as the reference anchor point for noise reduction.

[0026] Feedforward microphone (M1): Located on the outside of the headphones, facing the environment, it mainly picks up high-frequency environmental noise and wind noise in the range of 2000Hz to 10000Hz, and is used to capture instantaneously changing sound field information.

[0027] Voice microphone (M2): Located at the bottom of the ear stem, it uses beamforming to point towards the mouth and specifically extracts the voice-sensitive frequency band of 500Hz-4000Hz to improve the signal-to-noise ratio.

[0028] Differential pressure / accelerometer (M3): Non-acoustic sensor. Used to collect airflow pressure difference or shell vibration acceleration as a physical basis for distinguishing between "sound pressure" and "wind pressure / mechanical vibration", it is a key auxiliary input for solving motion wind noise.

[0029] This enables frequency band complementarity: By physically dividing the functions of M0 and M2, the contradiction that a single microphone cannot simultaneously address low-frequency sealing (inner ear) and high-frequency environmental perception (outer ear) is resolved.

[0030] Furthermore, it improves the applicability to motion: the introduction of M3 (pressure difference / acceleration) incorporates non-acoustic signals into acoustic processing, effectively distinguishing between "wind pressure impact" and "human voice impact," which is the key physical basis for solving wind noise during high-speed motion.

[0031] Specifically, the filtering unit is configured to perform the following operations: In scenarios where there is no phone call or media playback, an attenuation-type frequency band mask is generated to suppress environmental noise energy to below a preset threshold η(f). In a call scenario, A protective frequency band mask is generated by combining the speech presence probability pvad(f,t). Assign a gain weight κ=1 to the warm-tone frequency band where the human voice is located. Assign a decay weight κ<1 to non-voice audio segments.

[0032] This achieves both privacy and energy saving. In non-call mode, non-target sounds such as ambient human voices are actively suppressed through frequency band masking, which protects user privacy and reduces the computing load of the backend DSP. In addition, it can also ensure the preservation of human voice. In call mode, it uses κ weight to distinguish between human voice and non-human voice, avoiding the problem of traditional noise reduction algorithms "falsely killing" consonants.

[0033] Specifically, the color band filter constructed by the sound-to-color mapping module includes the following mapping relationships: Warm-Band: For frequencies between 200Hz and 800Hz, the hue is set to red / orange to identify the fundamental frequency of the human voice and chest resonance, and is given the highest protection priority in filtering. Mid-Vowel Band: Corresponding to 800Hz—2500Hz, the hue is set to yellow / green to identify vowel formants, allowing for moderate noise reduction in filtering; Air-Band (Air-feel / Wind Noise Band): Corresponding to 2500Hz-8000Hz, the hue is set to cyan / blue to identify sensitive areas for friction noise and wind noise. When the wind noise probability W(t) exceeds the standard, the gain weight of that band is automatically reduced.

[0034] Mapping sound signals to color images: Hue (H): strictly linearly mapped to the frequency axis; Low frequencies (200Hz) are defined as red, mid frequencies (1kHz) as green, and high frequencies (8kHz) as blue. This mapping aligns with human psychoacoustic perception that "warm colors = deep, cool colors = bright." Luminance (L): The logarithmic energy of the frequency point, log(1+αP(f,t)). The higher the energy, the brighter the color.

[0035] Opacity (Alpha, α): Mapped to the reliability of the signal.

[0036] When the wind noise probability W(t) is high, the blue pixels in the corresponding high-frequency region will become transparent (the alpha value decreases), indicating that the data in this region is unreliable in subsequent algorithms and the gain weight needs to be reduced.

[0037] By converting sound into color, engineers can visually see the distribution of wind noise (usually manifested as bright blue patches) during the development phase, thereby precisely adjusting filter parameters and greatly improving the feasibility of system debugging. In addition to its application in the consumer market, this design system also allows for different speaker combinations and housing pairings to produce varying effects during the R&D phase. This means that just because a speaker can achieve a certain performance level doesn't guarantee the entire wireless earphone will. This design system can serve as a test model for the design, thereby improving the R&D progress. Among them, color mapping conforms to human experience of perceiving the warmth or coolness of sound, making the filtering strategy more in line with the characteristics of human hearing, that is, softer and reducing the sharp frequency bands in the sound wave.

[0038] Specifically, the wind noise suppression module is configured as follows: By utilizing the coherence drop characteristics of the feedforward microphone (M1) and the feedback microphone (M0) in the high-frequency band, and the high-frequency energy surge characteristics of the feedforward microphone (M1), The wind noise probability W(t) is calculated by combining the vibration signal from the differential pressure sensor. When W(t) is greater than the preset threshold, the wind noise protection mechanism is triggered. This includes freezing the noise floor update, shifting the inflection point frequency of the high-pass filter from 3000Hz to 5000Hz, and reducing the brightness value of the airy / wind noise band in the color spectrogram.

[0039] in, The filtering unit is the logical center of the system. It does not process specific frequencies, but generates a macroscopic mask MaskFU1(f,t) based on the scene.

[0040] Non-call / non-playback mode (ambient monitoring / standby): The system is in privacy protection mode. FU1 generates an attenuation mask to suppress ambient noise energy to below a preset threshold η(f) (e.g., -40dB), effectively blocking surrounding conversations and retaining only the warning tone.

[0041] Wind noise in motion scenarios is the biggest challenge. This module utilizes a fusion of "sound and color images" and "physical sensors" to determine this. Decision logic: When the coherence of M1 and M0 in the high-frequency band suddenly drops (indicating that the two sounds are inconsistent, possibly due to airflow impacting M1 rather than the sound source), and the high-frequency energy of M1 rises sharply as a bright blue patch, while M3 detects a large low-frequency vibration, it is determined to be a wind noise event, and the wind noise probability W(t) is output.

[0042] Dynamic response: Once W(t)>0.8, the system immediately freezes the noise floor update (to prevent wind noise from polluting the reference noise floor) and moves the inflection point frequency of the high-pass filter from 3000Hz to 5000Hz, sacrificing some high-frequency detail in exchange for speech intelligibility, avoiding "pop-up" or "suffocation" during cycling or running.

[0043] Of course, specifically, multiple microphones also have the advantage of adjusting the speaker's vibration frequency based on the existing negative pressure in the ear canal, thereby reducing the negative pressure in the ear canal and improving wearing comfort through algorithms (that is, it can improve comfort while retaining the existing structure of in-ear headphones, and of course, it can also increase bandwidth and reduce sound leakage for non-in-ear headphones). Call mode: The system enters voice preservation mode.

[0044] FU1 combines the speech presence probability pvad(f,t) to assign an all-pass gain (κ=1) to the warm color band (200-800Hz) and an attenuation weight (κ<1) to the non-speech audio band (such as steady-state air conditioner sound), achieving the effect of "only hearing human voices and blocking the background".

[0045] Input accuracy is guaranteed: By "freezing the noise floor," wind noise peaks are prevented from polluting long-term noise estimates, ensuring that the noise reduction curve can quickly return to normal after a gust of wind.

[0046] Further eliminating the "breathing sensation," by dynamically adjusting the high-pass inflection point (3k->5k), more high-frequency human voice components are retained when wind noise is severe, avoiding muffled speech (mask effect) caused by excessive high-frequency cutoff, and significantly improving the accuracy of voice input in motion scenarios.

[0047] Specifically, the frequency-varying gain function calculation formula used by the noise reduction unit is as follows: ; in, Gc(f,t) is a color spectrogram; The fundamental frequency response curve is determined by the hue range of C(f,t). WindGain(W(t),f) is the probability of wind noise; W(t) represents the dynamically changing wind noise suppression gain.

[0048] Specifically, Gc(f,t) is determined by the hue range of the color spectrogram.

[0049] For example, when red / orange (Warm-Band) is detected, the gain is 1; When cyan / blue (airborne / wind noise band) is detected and there is no speech, the gain decreases linearly to 0.2~0.5; WindGain: A dedicated curve for suppressing wind noise.

[0050] As W(t) increases, the function exhibits exponential decay in the high-frequency range (>3kHz) while remaining stable in the low-frequency range, ensuring that wind noise is cut off while the fundamental frequency of human voice is not damaged.

[0051] Furthermore, through algorithmic fusion, the formula cleverly integrates traditional digital signal processing (DSP) mask (MaskFU1), innovative color perception weights (Gc), and motion sensor data (WindGain) into a single mathematical model.

[0052] This ensures a smooth audio transition. Since W(t) is a probability value (0-1), the gain change is continuous, avoiding sudden volume changes or clicking sounds caused by wind noise, thus ensuring auditory comfort.

[0053] A noise reduction method for multi-microphone smart sports headphones, based on the aforementioned multi-microphone smart sports headphone noise reduction system, includes the following steps: S1 acquires multiple audio and motion signals and performs time-frequency conversion; S2, based on the current operating scenario of the device, execute the first filtering strategy and generate a frequency band mask to isolate non-call sounds; S3 maps the time-frequency signal into a color spectrogram that includes hue, brightness, and opacity; S4, analyze the blue / cyan band energy and sensor signal in the color spectrogram to estimate the wind noise probability; S5. Based on the physiological perception weights and wind noise probability corresponding to the color bands, the final frequency response mask is synthesized. S6, the original signal is weighted and reconstructed using the final frequency response mask.

[0054] Complex multimodal signal processing is abstracted into standardized steps S1-S6. The "sound-to-color conversion + scene filtering" process avoids reliance on high-performance AI models. Through steps S3 and S4, "color mapping" and "color analysis" can be applied not only to the headphone end, but also to the headphone's research and development, debugging, and parameter calibration, making it a better parameter model.

[0055] By synthesizing a frequency response mask using the probability value W(t) in step S5, it is ensured that the gain adjustment is a continuous function rather than a step abrupt change. This eliminates the "clicking" or "fluctuating volume" sound commonly heard in wind noise processing and improves audio stability in motion scenarios.

[0056] Specifically, after step S3 and before step S4, Also includes: In response to a user's manual configuration operation on the graphical user interface (GUI), the color band parameters in the color spectrogram Gc(f,t) are remapped; The manual configuration operation includes: Adjust the hue interval boundaries of the warm color band, mid-frequency vowel band, or airy / wind noise band, or adjust the target gain weight Gc′(f,t) corresponding to a specific hue interval. Replace the original gain weight Gc(f,t) with the gain weight Gc′(f,t) after user remapping. Participate in the synthesis of the final frequency response mask described in step S5.

[0057] By leveraging the intuitiveness of "sound-to-color conversion," users no longer need to adjust the obscure EQ (equalizer), but instead adjust "red area (human voice) louder" or "blue area (wind noise) darker."

[0058] It's not just a simple UI control, but a fusion of visual aesthetics and auditory perception; users adjust the color to adjust the frequency response curve, and the noise cancellation adjustment method is a major highlight that distinguishes it from traditional headphone apps, greatly increasing the creativity of the user experience; For example, some audio enthusiasts have reduced their reliance on professional analysis and increased the playability of self-testing or inspection.

[0059] Specifically, step S3, which maps the time-frequency signal to a color spectrogram, is also applied to automated production line testing, including: Under a preset standard sound field environment, the initial frequency response data of the microphone array is collected, and the corresponding reference color spectrogram Cref(f,t) is generated. Calculate the deviation ΔE between the current color spectrogram Gc(f,t) of the earphone under test and the reference spectrogram within the preset color band; When the deviation value ΔE exceeds the preset tolerance ϵ, compensation filter parameters are automatically generated and embedded into the noise reduction unit of the earphone to compensate for hardware defects such as housing assembly tolerances or speaker frequency response differences. This solution can also be applied to earphone testing, especially for existing automated production processes, such as full-line inspection.

[0060] Figure 2This is a schematic diagram of the conversion of a time-domain audio signal into a time-frequency domain color spectrogram. That is, the band diagram can be combined with multiple audio signals to achieve the conversion. The middle part of the input segment can be understood as the use of multiple microphones as color bands for filtering. Figure 3 This is a schematic diagram of the color spectrum in the time-frequency domain after passing through the filtering unit, thus achieving a smooth transition between bands.

[0061] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A multi-microphone intelligent sports headphone noise reduction system, characterized in that, include: An audio acquisition module is provided with at least three microphones, which are used to acquire sound in the ear canal, ambient sound, and motion vibration signals, respectively. The preprocessing module is used to synchronize and calibrate the output signal of the audio acquisition module; A filtering unit is used to generate a frequency band mask MaskFU1(f,t) based on the scene recognition results. The filtering unit filters out non-target sounds in non-call scenarios, and the scene recognition results include call activation status, media playback status, and motion posture. The sound-to-color mapping module is used to convert time-domain audio signals into time-frequency domain color spectrograms Gc(f,t). This also includes mapping frequency to hue H, energy to brightness L, and mapping the probability of speech presence or wind noise to opacity α. A wind noise suppression module is used to calculate the wind noise probability W(t) based on the color features output by the sound-to-color mapping module and the motion vibration signal, and to dynamically adjust the high-frequency attenuation factor. A noise reduction unit is used to filter the input signal by combining the frequency band mask MaskFU1(f,t), the color spectrogram Gc(f,t) and the wind noise probability W(t) through the frequency-varying gain function Hfinal(f,t); An audio reconstruction module is used to convert the filtered signal into an analog audio output.

2. The multi-microphone intelligent sports headphone noise reduction system according to claim 1, characterized in that, The audio acquisition module includes: Feedback microphone (M0), which is located inside the ear canal and mainly covers the mid-low frequency band of 300Hz-3500Hz; A feedforward microphone (M1) is located on the outside of the earphone and mainly covers the high frequency band of 2000Hz-10000Hz. A voice microphone (M2) is located on the side of the ear stem near the mouth, and mainly covers the human voice sensitive frequency band of 500Hz-4000Hz; A differential pressure or acceleration sensor (M3) is used to collect airflow pressure difference or housing vibration signals as an auxiliary feature for wind noise detection.

3. The multi-microphone intelligent sports headphone noise reduction system according to claim 2, characterized in that, The filtering unit is configured to perform the following operations: In scenarios where there is no phone call or media playback, an attenuation-type frequency band mask is generated to suppress environmental noise energy to below a preset threshold η(f). In a call scenario, A protective frequency band mask is generated by combining the speech presence probability pvad(f,t). Assign a gain weight κ=1 to the warm-tone frequency band where the human voice is located. Assign a decay weight κ<1 to non-voice audio segments.

4. The multi-microphone intelligent sports headphone noise reduction system according to claim 3, characterized in that, The color band filter constructed by the sound-to-color mapping module includes the following mapping relationships: Warm color band: For frequencies between 200Hz and 800Hz, the hue is set to red / orange to identify the fundamental frequency of the human voice and chest resonance, and is given the highest protection priority in filtering. Mid-frequency vowel band: Corresponding to 800Hz—2500Hz, the hue is set to yellow / green to identify vowel formants, allowing for moderate noise reduction in filtering; Airiness / Wind Noise Band: Corresponding to 2500Hz-8000Hz, the hue is set to cyan / blue to identify sensitive areas for friction noise and wind noise. When the wind noise probability W(t) exceeds the standard, the gain weight of that band is automatically reduced.

5. The multi-microphone intelligent sports headphone noise reduction system according to claim 4, characterized in that, The wind noise suppression module is configured as follows: By utilizing the coherence drop characteristics of the feedforward microphone (M1) and the feedback microphone (M0) in the high-frequency band, and the high-frequency energy surge characteristics of the feedforward microphone (M1), The wind noise probability W(t) is calculated by combining the vibration signal from the differential pressure sensor. When W(t) is greater than the preset threshold, the wind noise protection mechanism is triggered. This includes freezing the noise floor update, shifting the inflection point frequency of the high-pass filter from 3000Hz to 5000Hz, and reducing the brightness value of the airy / wind noise band in the color spectrogram.

6. The multi-microphone intelligent sports headphone noise reduction system according to claim 5, characterized in that, The frequency-varying gain function calculation formula used by the noise reduction unit is as follows: ； in, Gc(f,t) is a color spectrogram; The fundamental frequency response curve is determined by the hue range of C(f,t). WindGain(W(t),f) is the probability of wind noise; W(t) represents the dynamically changing wind noise suppression gain.

7. A noise reduction method for multi-microphone smart sports headphones, based on the multi-microphone smart sports headphone noise reduction system according to any one of claims 1-6, characterized in that, Includes the following steps: S1 acquires multiple audio and motion signals and performs time-frequency conversion; S2, based on the current operating scenario of the device, execute the first filtering strategy and generate a frequency band mask to isolate non-call sounds; S3 maps the time-frequency signal into a color spectrogram that includes hue, brightness, and opacity; S4, analyze the blue / cyan band energy and sensor signal in the color spectrogram to estimate the wind noise probability; S5. Based on the physiological perception weights and wind noise probability corresponding to the color bands, the final frequency response mask is synthesized. S6, the original signal is weighted and reconstructed using the final frequency response mask.

8. The noise reduction method for multi-microphone smart sports headphones according to claim 7, characterized in that, After S3 and before S4, Also includes: In response to a user's manual configuration operation on the graphical user interface, the color band parameters in the color spectrogram Gc(f,t) are remapped; The manual configuration operation includes: Adjust the hue interval boundaries of the warm color band, mid-frequency vowel band, or airy / wind noise band, or adjust the target gain weight Gc′(f,t) corresponding to a specific hue interval. Replace the original gain weight Gc(f,t) with the gain weight Gc′(f,t) after user remapping. Participate in the synthesis of the final frequency response mask described in step S5.

9. The noise reduction method for multi-microphone smart sports headphones according to claim 7, characterized in that, The S3 method, which maps time-frequency signals into color spectrograms, is also applied to automated production line testing, including: Under a preset standard sound field environment, the initial frequency response data of the microphone array is collected, and the corresponding reference color spectrogram Cref(f,t) is generated. Calculate the deviation ΔE between the current color spectrogram Gc(f,t) of the earphone under test and the reference spectrogram within the preset color band; When the deviation value ΔE exceeds the preset tolerance ϵ, compensation filter parameters are automatically generated and the compensation filter parameters are fixed into the noise reduction unit of the earphone to achieve hardware defect compensation for housing assembly tolerance or speaker frequency response difference.

Images