Bone conduction splicing noise reduction method, audio device and computer readable storage medium

Abstract

The application discloses a bone air conduction splicing noise reduction method, an audio device and a computer readable storage medium, relates to the technical field of signal processing, and comprises the following steps: acquiring an air conduction audio signal, a bone conduction audio signal and a pre-trained signal noise reduction model; extracting a first component signal belonging to a first frequency band from the bone conduction audio signal, and extracting a second component signal belonging to a second frequency band from the air conduction audio signal, wherein the first frequency band and the second frequency band respectively belong to different frequency band intervals in the frequency domain, and the upper limit value of the first frequency band is less than the upper limit value of the second frequency band; performing frequency domain merging on the first component signal and the second component signal to generate a spliced audio signal; and inputting the spliced audio signal into the signal noise reduction model to obtain a target audio signal after noise reduction. The application improves the signal noise reduction effect.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, and in particular to a bone-air conduction splicing noise reduction method, an audio device, and a computer-readable storage medium. Background Technology

[0002] With the increasing popularity of wearable audio devices, users' demands for voice call quality in complex acoustic environments are rising. Deep learning-based voice noise reduction technology has become the mainstream solution for improving call clarity, among which the method of integrating multiple sensors (such as air conduction microphones and bone conduction microphones) has received widespread attention.

[0003] Currently, deep learning-based speech denoising methods primarily use noisy speech signals acquired by air conduction microphones as input and clean speech signals as the learning target, estimating the denoised target signal through a neural network. This pure air conduction scheme performs well in scenarios with high signal-to-noise ratios, effectively preserving the naturalness of the speech. Meanwhile, some schemes introduce bone conduction microphones, which are naturally immune to environmental noise, as auxiliary sensors. These microphones input both bone conduction and air conduction audio signals into the network, aiming to leverage the high signal-to-noise ratio of bone conduction signals under strong noise conditions to assist air conduction denoising.

[0004] However, all of the above solutions perform poorly in noisy environments. In the pure air conduction scheme, the low-frequency portion of the air conduction audio signal is easily masked by noise, leading to a sharp deterioration in noise reduction. While the bone conduction fusion scheme has some noise resistance advantages, its prediction target is still the clean air conduction audio signal, failing to fully utilize the bone conduction signal to improve the overall noise reduction effect. Therefore, how to improve the signal noise reduction effect has become an urgent problem to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a bone-air conduction splicing noise reduction method, an audio device, and a computer-readable storage medium, aiming to solve the technical problem of how to improve the signal noise reduction effect.

[0006] To achieve the above objectives, this application provides a method for noise reduction through bone air conduction splicing, the method comprising the following steps: Acquire air conduction audio signals, bone conduction audio signals, and a pre-trained signal denoising model; The first component signal belonging to the first frequency band is extracted from the bone conduction audio signal, and the second component signal belonging to the second frequency band is extracted from the air conduction audio signal, wherein the first frequency band and the second frequency band belong to different frequency band intervals in the frequency domain, and the upper limit of the first frequency band is less than the upper limit of the second frequency band. The first component signal and the second component signal are frequency domain merged to generate a spliced ​​audio signal; The spliced ​​audio signal is input into the signal denoising model to obtain the denoised target audio signal.

[0007] In one embodiment, the step of frequency-domain merging of the first component signal and the second component signal to generate a spliced ​​audio signal includes: Within the transition frequency band between the first frequency band and the second frequency band, the frequency point values ​​of the first component signal and the second component signal are weighted and merged according to a preset weight, and the merging result is used as the value of the spliced ​​audio signal at the corresponding frequency point. Outside the transition frequency band between the first frequency band and the second frequency band, the frequency point value of the first component signal in the first frequency band and the frequency point value of the second component signal in the second frequency band are respectively used as the value of the spliced ​​audio signal at the corresponding frequency point.

[0008] In one embodiment, the step of weighting and merging the frequency values ​​of the first component signal and the second component signal according to a preset weight includes: The weights corresponding to the first component signal and the second component signal are determined respectively, wherein, from the low-frequency end to the high-frequency end of the transition frequency band, the weight of the first component signal decreases and the weight of the second component signal increases. The frequency values ​​of the first component signal and the second component signal are weighted and combined according to the determined weights.

[0009] In one embodiment, the method for determining the weights corresponding to the first component signal and the second component signal respectively further includes: Calculate the local signal-to-noise ratio of the air-conducted audio signal within the transition frequency band; The initial weights corresponding to the first component and the second component are determined based on the local signal-to-noise ratio, wherein the higher the local signal-to-noise ratio, the higher the initial weight of the second component and the lower the initial weight of the first component. Based on the initial weights, the weights corresponding to the first component are increased sequentially according to a preset step size, and the weights corresponding to the second component are decreased sequentially.

[0010] In one embodiment, after the steps of acquiring the air conduction audio signal, the bone conduction audio signal, and the pre-trained signal denoising model, the method further includes: Determine the global signal-to-noise ratio of the air-conducted audio signal; When the global signal-to-noise ratio is lower than a preset signal-to-noise ratio threshold, the upper limit frequency of the first frequency band is extended to expand the bandwidth of the transition band. When the global signal-to-noise ratio is higher than the preset signal-to-noise ratio threshold, the upper limit frequency of the first frequency band is reduced to decrease the bandwidth of the transition frequency band.

[0011] In one embodiment, the spliced ​​audio signal includes an amplitude spectrum and a phase spectrum, and the step of inputting the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal includes: The amplitude spectrum of the spliced ​​audio signal is input into the signal denoising model to predict the denoised amplitude spectrum; The amplitude spectrum after noise reduction is synthesized with the phase spectrum of the spliced ​​audio signal, and the noise-reduced target audio signal is output after time-domain transformation.

[0012] In one embodiment, the step of inputting the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal includes: The spliced ​​audio signal is split into multiple sub-band signals; Each of the sub-band signals is input into the signal denoising model for denoising processing to obtain the denoised sub-band signal corresponding to each of the sub-band signals. The denoised sub-band signals are synthesized to obtain the denoised target audio signal.

[0013] In one embodiment, the step of splitting the spliced ​​audio signal into multiple sub-band signals includes: The boundary frequencies of multiple critical frequency bands are determined based on the characteristics of human hearing, wherein the bandwidth of the critical frequency band increases with the increase of the center frequency. According to the boundary frequencies of each of the critical frequency bands, the spliced ​​audio signal is divided into multiple sub-band signals that correspond one-to-one with each of the critical frequency bands.

[0014] Furthermore, to achieve the above objectives, this application also provides an audio device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps of the bone conduction splicing noise reduction method described above. In addition, to achieve the above objectives, this application also provides a readable storage medium, which is a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to implement the steps of the bone air conduction splicing noise reduction method as described above.

[0015] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the bone conduction splicing noise reduction method described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: This application embodiment extracts a low-frequency first component signal from the bone conduction audio signal and a high-frequency second component signal from the air conduction audio signal. This fully utilizes the high signal-to-noise ratio of the bone conduction signal in the low-frequency band to avoid the problem of air conduction low frequencies being masked by noise, while preserving the speech harmonics and naturalness information of the air conduction audio signal in the mid-to-high frequency bands. These two component signals, belonging to different frequency bands and each with its own advantages, are then frequency-domain merged. The resulting spliced ​​audio signal relies on bone conduction to ensure noise immunity in the low-frequency band and on air conduction to maintain speech naturalness in the high-frequency band. Thus, at the input level, a signal with both noise immunity and acoustic integrity is constructed. The model expresses the integral fusion features. After inputting the spliced ​​audio signal into the noise reduction model, the model processes the high-quality input after frequency band optimization and reconstruction. Since the low-frequency component originates from bone conduction and has a high signal-to-noise ratio, the network does not need to perform significant noise suppression in this frequency band, thus avoiding the problem of low-frequency performance degradation in the pure air conduction scheme. At the same time, the high-frequency component of air conduction is completely preserved, so that the output audio can still maintain good clarity and naturalness. Finally, through frequency band division and selective merging, the scheme enables bone conduction to substantially dominate low-frequency noise reduction and air conduction to be responsible for high-frequency quality preservation. The two complement each other, thereby improving the overall signal noise reduction effect. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of the bone air conduction splicing noise reduction method of this application; Figure 2 This is a schematic diagram of the signal noise reduction process involved in an embodiment of the bone air conduction splicing noise reduction method of this application; Figure 3 This is a flowchart illustrating the second embodiment of the bone air conduction splicing noise reduction method of this application; Figure 4 This is another flowchart illustrating the second embodiment of the bone air conduction splicing noise reduction method of this application; Figure 5 This is a schematic diagram of the device structure according to an embodiment of the audio device of this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the technical solutions of 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] With the continuous development of deep learning technology and the increasing popularity of wearable audio devices (such as smart headphones, AR / VR glasses, etc.), deep learning-based speech denoising algorithms have become a key technology for improving the quality of voice calls. Traditional speech denoising methods typically use audio signals collected by air conduction microphones as the processing target, and obtain clean near-end audio signals through model processing.

[0023] Air conduction audio signals offer high speech fidelity, but they are easily submerged in noise, leading to a significant decrease in signal-to-noise ratio (SNR). Bone conduction signals, while offering relatively lower speech fidelity, are less sensitive to environmental noise and maintain a high SNR even in noisy environments. With the increasing prevalence of bone conduction microphones in wearable audio devices, effectively fusing bone conduction and air conduction audio signals has become a hot research topic in the field of speech noise reduction.

[0024] Currently, the mainstream speech noise reduction architectures mainly include the following two types: The first category is pure air conduction speech denoising technology. This architecture uses noisy air conduction audio signals as model input and clean air conduction audio signals as prediction targets. In high signal-to-noise ratio scenarios, this method can preserve speech information in the air conduction microphone well, and the denoising effect is good; however, in low signal-to-noise ratio scenarios, the speech denoising effect is significantly reduced because it relies solely on the air conduction microphone signal.

[0025] The second category is speech denoising techniques that fuse bone conduction and air conduction audio signals. This architecture simultaneously inputs noisy air conduction audio signals and bone conduction signals, with the model still targeting the clean air conduction audio signal. In low signal-to-noise ratio (SNR) scenarios, this method leverages the noise insensitivity of bone conduction signals, resulting in a significant improvement in denoising performance compared to pure air conduction methods. However, when noise levels are extremely high, the low-frequency components of the air conduction audio signal are often completely submerged, making it difficult to effectively recover the low-frequency components even with the introduction of bone conduction signals. In high SNR scenarios, this method maximizes the preservation of air conduction audio signals while also utilizing bone conduction information for assistance, thus achieving better denoising results.

[0026] Therefore, although the two architectures mentioned above have different input information, they share the common feature of using clean air-conducted audio signals as the sole prediction target. This setting leads to difficulties in recovering the low-frequency components of air-conducted audio signals effectively using existing models in low signal-to-noise ratio scenarios when the air-conducted audio signals are severely submerged by noise, thus limiting further improvements in speech denoising performance.

[0027] To address the aforementioned issues, the main solution of this application is as follows: acquiring air conduction audio signals, bone conduction audio signals, and a pre-trained signal denoising model; extracting a first component signal belonging to a first frequency band from the bone conduction audio signal, and extracting a second component signal belonging to a second frequency band from the air conduction audio signal, wherein the first frequency band and the second frequency band belong to different frequency band intervals in the frequency domain, and the upper limit of the first frequency band is less than the upper limit of the second frequency band; merging the first component signal and the second component signal in the frequency domain to generate a spliced ​​audio signal; and inputting the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal.

[0028] This application extracts a low-frequency first component signal from the bone conduction audio signal and a high-frequency second component signal from the air conduction audio signal. This fully utilizes the high signal-to-noise ratio of the bone conduction signal in the low-frequency band to avoid the problem of noise masking in the low-frequency air conduction signal, while preserving the speech harmonics and naturalness information of the air conduction audio signal in the mid-to-high frequency bands. The two component signals, belonging to different frequency bands and each with its own advantages, are then frequency-domain merged. The resulting spliced ​​audio signal relies on bone conduction to ensure noise immunity in the low-frequency band and on air conduction to maintain speech naturalness in the high-frequency band. Thus, at the input level, a signal with both noise immunity and acoustic integrity is constructed. The fusion feature expression is used. After the spliced ​​audio signal is input into the noise reduction model, the model processes the high-quality input after frequency band optimization and reconstruction. Since the low-frequency component comes from bone conduction and has a high signal-to-noise ratio, the network does not need to perform significant noise suppression in this frequency band, thus avoiding the problem of low-frequency performance degradation in the pure air conduction scheme. At the same time, the high-frequency component of air conduction is completely preserved, so that the output audio can still maintain good clarity and naturalness. Finally, through frequency band division and selective merging, the bone conduction substantially dominates low-frequency noise reduction, while the air conduction is responsible for maintaining high-frequency quality. The two complement each other, thereby improving the overall signal noise reduction effect.

[0029] Based on this, this application proposes a first embodiment of a bone air conduction splicing noise reduction method. In this embodiment, referring to... Figure 1 As shown, the bone air conduction splicing noise reduction method includes the following steps S10~S40: Step S10: Acquire air conduction audio signal, bone conduction audio signal, and pre-trained signal denoising model; The air conduction audio signal is acquired by an air conduction microphone, which obtains the speech signal by picking up sound waves propagating in the air. It can preserve the harmonic structure and naturalness characteristics of speech relatively completely, but it is easily affected by environmental noise.

[0030] Bone conduction audio signals are acquired by a bone conduction microphone, which obtains speech information by picking up vibration signals transmitted by tissues such as the skull when the speaker speaks. It has a natural ability to suppress ambient airborne noise and has a significant signal-to-noise ratio advantage in the low-frequency band.

[0031] The signal denoising model can be a neural network model that has been pre-trained based on deep learning algorithms. It is used to denoise the input audio signal and output the denoised target audio.

[0032] Step S20: Extract a first component signal belonging to the first frequency band from the bone conduction audio signal, and extract a second component signal belonging to the second frequency band from the air conduction audio signal, wherein the first frequency band and the second frequency band belong to different frequency band intervals in the frequency domain, and the upper limit of the first frequency band is less than the upper limit of the second frequency band. Component signals can be extracted using filtering operations in digital signal processing. For example, bandpass filters can be used to filter the frequency bands of bone conduction and air conduction audio signals separately. The first frequency band is a relatively low frequency range. Taking advantage of the high signal-to-noise ratio of bone conduction signals in this band, low-frequency components with excellent noise immunity are provided for subsequent splicing. The second frequency band is a relatively high frequency range. Taking advantage of the rich speech details and naturalness information retained by air conduction audio signals in this band, high-frequency components with good auditory perception quality are provided for subsequent splicing.

[0033] Step S30: Combine the first component signal and the second component signal in the frequency domain to generate a spliced ​​audio signal; The first and second component signals can be converted to the frequency domain using Fourier transform to obtain their respective spectral data. Then, the two spectral data are concatenated, combining the low-frequency spectrum of the first component signal with the high-frequency spectrum of the second component signal in the frequency domain to form a complete spectrum, creating a fused spectrum containing both low-frequency bone conduction and high-frequency air conduction information. Finally, an inverse Fourier transform is performed on the fused spectrum to convert it back to the time domain, thus obtaining the concatenated audio signal.

[0034] Step S40: Input the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal.

[0035] The generated spliced ​​audio signal is used as input features and fed into a pre-trained signal denoising model. The model performs end-to-end denoising mapping on the input spliced ​​audio signal through forward propagation calculation of the neural network, and finally outputs the denoised target audio signal.

[0036] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description and will not be repeated hereafter. Based on this, the step of frequency domain merging the first component signal and the second component signal to generate a spliced ​​audio signal includes: Step A10: In the transition frequency band between the first frequency band and the second frequency band, the frequency point values ​​of the first component signal and the second component signal are weighted and merged according to a preset weight, and the merging result is used as the value of the spliced ​​audio signal at the corresponding frequency point. The preset frequency threshold is a pre-defined frequency domain boundary point used to define the transition area between the first frequency band and the second frequency band. This transition frequency band is located between the upper limit of the first frequency band and the lower limit of the second frequency band, and its bandwidth can be configured according to the actual application scenario and signal characteristics.

[0037] In the transition frequency band, in order to avoid phase abrupt changes and spectral leakage caused by spectral discontinuity at the frequency domain splicing point between the first component signal and the second component signal, and thus avoid introducing perceptible distortion or noise, this embodiment adopts a weighted merging method for processing.

[0038] Specifically, for each frequency point within the transition band, the amplitude or energy value of the first component signal and the amplitude or energy value of the second component signal at that frequency point are obtained respectively. Then, the two signal values ​​are weighted and summed according to a preset weight, and the weighted merging result is used as the final value of the spliced ​​audio signal at that frequency point.

[0039] The preset weights can be dynamically determined based on the relative position of the frequency points within the transition frequency band. For example, a gradual weight can be used so that the weight of the first component signal is larger and the weight of the second component signal is smaller when it is closer to the first frequency band, and the opposite is true when it is closer to the second frequency band. This achieves a smooth connection between the two signals in the frequency domain, maintaining the noise reduction advantage of the low-frequency bone conduction signal and the naturalness advantage of the high-frequency air conduction audio signal, while avoiding sound quality degradation caused by abrupt changes in the frequency domain.

[0040] Step A20: Outside the transition frequency band between the first frequency band and the second frequency band, the frequency point value of the first component signal in the first frequency band and the frequency point value of the second component signal in the second frequency band are respectively used as the value of the spliced ​​audio signal at the corresponding frequency point.

[0041] Beyond the transition band, there are low-frequency regions below the transition band start frequency and high-frequency regions above the transition band cutoff frequency.

[0042] In the low-frequency region below the transition band start frequency, which belongs to the first frequency band, the value of the first component signal corresponding to the frequency point in this region is directly used as the value of the spliced ​​audio signal at that frequency point; in the high-frequency region above the transition band cutoff frequency, which belongs to the second frequency band, the value of the second component signal corresponding to the frequency point in this region is directly used as the value of the spliced ​​audio signal at that frequency point.

[0043] This processing method fully preserves the noise immunity of bone conduction signals in the low-frequency band and the rich detail of air conduction audio signals in the high-frequency band outside the transition band. At the same time, together with the weighted merging results in the transition band, it constitutes a complete spliced ​​audio signal spectrum, ensuring that the spliced ​​audio signal has good noise immunity and natural speech in the entire frequency range, while maintaining continuity and smoothness in the frequency domain.

[0044] In one possible implementation, the step of weighting and merging the frequency values ​​of the first component signal and the second component signal according to a preset weight includes: Step B10: Determine the weights corresponding to the first component signal and the second component signal respectively, wherein, from the low-frequency end to the high-frequency end of the transition band, the weight of the first component signal decreases and the weight of the second component signal increases. The transition band has a defined frequency range, with its low-frequency end being the starting frequency closer to the first frequency band and its high-frequency end being the cutoff frequency closer to the second frequency band. Within the transition band, it is necessary to determine the weight values ​​for the first and second component signals for each frequency point, in order to use in subsequent weighted combining calculations.

[0045] Specifically, during the transition from the low-frequency end to the high-frequency end of the transition band, the weight corresponding to the first component signal decreases, while the weight corresponding to the second component signal increases. This weight change can be linear, for example, the weight can be considered a linear function of frequency, with the first weight decreasing linearly and the second weight increasing linearly as the frequency increases uniformly from the low-frequency end to the high-frequency end; or it can be nonlinear, such as an S-shaped curve or an exponential curve, to make the weight change more gradual or more rapid in certain frequency bands, adapting to the splicing smoothness requirements in different acoustic scenarios.

[0046] Furthermore, at the low-frequency start of the transition band, the weight of the first component signal can be set to 1, and the weight of the second component signal can be set to 0, so that the spliced ​​audio signal is completely dominated by the bone conduction signal at this position. At the high-frequency cutoff of the transition band, the weight of the first component signal can be set to 0, and the weight of the second component signal can be set to 1, so that the spliced ​​audio signal is completely dominated by the air conduction audio signal at this position. In the middle of the transition band, the weights of the two signals are each 0.5, achieving proportional fusion, thereby forming a gradual transition effect from bone conduction dominance to air conduction dominance in the frequency domain.

[0047] Step B20: The frequency values ​​of the first component signal and the second component signal are weighted and merged according to the determined weights.

[0048] For each frequency point within the transition band, after the weights of the first and second component signals corresponding to that frequency point have been determined in step B10, the frequency point value of the first component signal at that frequency point is multiplied by its corresponding first weight, and the frequency point value of the second component signal at that frequency point is multiplied by its corresponding second weight. Then, the products of the two are added together to obtain the weighted and merged frequency point value, which is used as the final value of the spliced ​​audio signal at that frequency point.

[0049] The weighted merging process can be expressed as the formula Y(f)=W1(f)·X1(f)+W2(f)·X2(f), where f is the current frequency point within the transition band, X1(f) and X2(f) are the values ​​of the first component signal and the second component signal at frequency point f, respectively, and W1(f) and W2(f) are the weights of the first component signal and the second component signal at frequency point f, respectively, and satisfy W1(f)+W2(f)=1 to ensure that the energy of the merged signal does not have an overall shift in amplitude.

[0050] Through the above weighted merging process, the bone conduction signal and air conduction audio signal are naturally integrated in the transition frequency band. Bone conduction is the main component at the low frequency end to ensure noise resistance, while air conduction is the main component at the high frequency end to maintain the naturalness of speech. The two signals are smoothly connected in the middle region through weight gradient, thereby avoiding the sound quality degradation caused by frequency domain abrupt changes. At the same time, the continuity and integrity of the spliced ​​audio signal in the entire frequency domain are guaranteed, providing high-quality input features for the subsequent signal denoising model.

[0051] In one possible implementation, the method further includes the step of determining the weights corresponding to the first component signal and the second component signal respectively: Step C10: Calculate the local signal-to-noise ratio of the air-conducted audio signal within the transition frequency band; Local signal-to-noise ratio (SNR) refers to the ratio of signal power to noise power in a specific frequency range within the transition band of an air-conducted audio signal. It characterizes the degree of contamination of the air-conducted audio signal in that frequency band. Specifically, speech activity detection technology can be used to distinguish between speech and non-speech segments in the air-conducted audio signal. In the non-speech segments, the noise power spectrum within the transition band can be estimated, and in the speech segments, the total power spectrum including both speech and noise can be calculated. The ratio of the total power spectrum to the noise power spectrum is then used as the local SNR. Alternatively, minimum tracking or noise estimation algorithms based on recursive averaging can be used to update the noise power spectrum in real time, thereby dynamically calculating the local SNR to adapt to the actual situation of rapid changes in SNR under strong noise environments.

[0052] Step C20: Determine the initial weights corresponding to the first component and the second component based on the local signal-to-noise ratio, wherein the higher the local signal-to-noise ratio, the higher the initial weight of the second component and the lower the initial weight of the first component. The local signal-to-noise ratio (SNR) directly reflects the reliability of the air-conducted audio signal within the transition band. When the local SNR of the air-conducted audio signal is high within the transition band, it indicates that the signal is less affected by noise in that frequency band and has better signal quality. In this case, the initial weight of the second component signal (derived from air conduction) is set higher, and the initial weight of the first component signal (derived from bone conduction) is set lower. This allows the stitching process to rely more on the air-conducted audio signal to contribute frequency values ​​in the transition band, fully utilizing the rich speech details and naturalness information preserved in the mid-to-high frequency range. Conversely, when the local SNR is low, it indicates that the air-conducted audio signal is severely contaminated by noise in that frequency band. In this case, the initial weight of the first component signal (derived from bone conduction) is set higher, and the initial weight of the second component signal is set lower. This leverages the natural immunity of the bone conduction signal to environmental noise to dominate the stitching of the transition band, preventing noise from being introduced into the stitched audio signal.

[0053] Specifically, the initial weights can be determined by a preset mapping function. For example, the local signal-to-noise ratio can be mapped to the [0,1] interval and used directly as the initial weight of the second component. The initial weight of the first component is 1 minus the mapping value, so that the sum of the initial weights of the two signals is always 1, maintaining the amplitude consistency of the merged signal.

[0054] Step C30: Based on the initial weights, the weights corresponding to the first component are increased sequentially according to a preset step size, and the weights corresponding to the second component are decreased sequentially.

[0055] The initial weights are static weight values ​​determined based on the local signal-to-noise ratio of the air-conducted audio signal within the transition band. They reflect the average signal quality within a single frequency band, but cannot fully reflect the differences in spectral distribution between different frequency points within the transition band. Therefore, further adjustments are made based on the initial weights.

[0056] Specifically, the process iterates through each frequency point from the low-frequency end to the high-frequency end of the transition band, gradually increasing the weight of the first component signal and gradually decreasing the weight of the second component signal according to a preset step size. It should be noted that this adjustment trend is consistent with the trend in step B10 where the first weight decreases and the second weight increases from the low-frequency end to the high-frequency end. In other words, it essentially adds an incremental adjustment related to the frequency position to the initial weight baseline, ensuring that the final weight reflects both the overall signal-to-noise ratio of the air-conducting audio signal within the transition band and the physical law of decreasing bone conduction signal advantage and increasing air conduction audio signal advantage from the low-frequency end to the high-frequency end. This results in a better balance between noise resistance and speech naturalness in the splicing result within the transition band, further improving the quality of the spliced ​​audio signal.

[0057] In one possible implementation, after the steps of acquiring the air conduction audio signal, the bone conduction audio signal, and the pre-trained signal denoising model, the method further includes: Step D10: Determine the global signal-to-noise ratio of the air conduction audio signal; Global signal-to-noise ratio (SNR) refers to the ratio of signal power to noise power of an air-conducted audio signal across the entire frequency band. It is used to characterize the degree to which an air-conducted audio signal is contaminated by environmental noise.

[0058] Specifically, speech activity detection technology can be used to distinguish between speech segments and non-speech segments in air-conducted audio signals. The noise power spectrum within the entire frequency band can be estimated in the non-speech segments, and the total power spectrum including speech and noise can be calculated in the speech segments. The ratio of the total power to the noise power in the entire frequency band can be used as the global signal-to-noise ratio.

[0059] Minimum tracking or noise estimation algorithms based on recursive averaging can also be used to track changes in the noise floor in real time across the entire frequency band, thereby dynamically calculating the global signal-to-noise ratio to adapt to rapid fluctuations in the signal-to-noise ratio under different noise environments.

[0060] Step D20: When the global signal-to-noise ratio is lower than a preset signal-to-noise ratio threshold, the upper limit frequency of the first frequency band is extended to expand the bandwidth of the transition band. The preset signal-to-noise ratio (SNR) threshold is a pre-defined global SNR threshold value used to determine whether the current acoustic environment is a noisy scene. When the global SNR is lower than this threshold, it indicates that the air-conducted audio signal is subject to relatively severe noise pollution across the entire frequency band, especially the problem of the low-frequency part of the air-conducted audio signal being masked by noise will be more prominent.

[0061] In this case, by extending the upper limit frequency of the first frequency band, the bone conduction signal can participate in signal splicing in a higher frequency range, which is equivalent to expanding the bandwidth of the transition band. That is, the starting frequency of the transition band remains unchanged while the cutoff frequency extends to a higher frequency, thereby allowing a larger range of frequency domains to fall within the transition band.

[0062] The expanded transition band increases the number of frequency points for weighted merging, allowing more frequency points to leverage the noise immunity of bone conduction signals in their respective frequency bands under strong noise conditions, further suppressing the impact of noise on the spliced ​​audio signal and effectively improving the noise reduction effect.

[0063] Step D30: When the global signal-to-noise ratio is higher than the preset signal-to-noise ratio threshold, the upper limit frequency of the first frequency band is reduced to decrease the bandwidth of the transition frequency band.

[0064] When the global signal-to-noise ratio is higher than the preset signal-to-noise ratio threshold, it indicates that the current acoustic environment is relatively quiet or the noise level is low. The air-conducted audio signal has good signal quality across the entire frequency band, and its low-frequency part is not severely masked by noise.

[0065] In this case, by shrinking the upper limit frequency of the first frequency band, the bone conduction signal can be cut off and participate in splicing in a lower frequency range, which is equivalent to reducing the bandwidth of the transition band. That is, the starting frequency of the transition band remains unchanged, while the cutoff frequency shrinks back to a lower frequency, so that fewer frequency points fall into the transition band, while a larger frequency domain is directly assigned to the second frequency band and the high frequency point value of the air conduction audio signal is used.

[0066] The reduced transition band decreases the number of frequency points in the weighted merging, allowing more frequency points to directly utilize the rich speech harmonic structure and naturalness information preserved by the air conduction audio signal under low noise conditions. This avoids affecting the natural auditory experience of speech due to excessive introduction of bone conduction signals, thus better maintaining speech quality while ensuring noise reduction.

[0067] Through the adaptive mechanism described above, which dynamically adjusts the upper limit frequency of the first frequency band and the width of the transition frequency band based on the global signal-to-noise ratio, the bone-air conduction splicing noise reduction method can adaptively adjust the fusion range of bone conduction signals and air conduction audio signals under different noise scenarios, realizing flexible switching between noise reduction priority under strong noise and fidelity priority under weak noise, further improving the robustness and adaptability of the overall noise reduction scheme.

[0068] For example, to aid in understanding the technical concept or principle of the bone air conduction splicing noise reduction method combined with the first embodiment described above, a specific embodiment is now listed. In this specific embodiment, refer to... Figure 2 As shown, the signal denoising process includes: Step 1: Extract the high-frequency portion of the air conduction audio signal and the low-frequency portion of the bone conduction signal; Step 2: The high-frequency component of the air conduction audio signal is spliced ​​with the low-frequency component of the bone conduction signal. The specific splicing method is as follows: Assuming the signal has a 16kHz sampling rate and a 512-point FFT is performed, the frequency domain dimension is 257, and the frequency range is 8kHz. Assuming the air conduction frequency domain characteristics are x1 = [T, 257] and the bone conduction frequency domain characteristics are x2 = [T, 257] The stitching is performed around 1K, which corresponds to a 32-dimensional frequency domain. To ensure the smoothness of the frequency domain characteristics of the stitched signal, a smooth stitching method is proposed, such as a linear smoothing strategy. x[T, :28]=x2[T, :28] (Frequencies below 28 are all bone conduction signals); x[T, 35:]=x1[T, 35:] (Frequency points above 35 are all air-conducted audio signals); Weighted average of bone conduction signals and low-frequency signals between 28 and 35 frequency points: x[T, 28:29]=0.125 x1[T, 28:29]+ 0.875 x2[T, 28:29]; x[T, 29:30]=0.25 x1[T, 29:30]+0.75 x2[T, 29:30]; x[T, 30:31]=0.375 x1[T, 30:31]+0.625 x2[T, 30:31]; x[T, 31:32]=0.5 x1[T, 31:32]+0.5 x2[T, 31:32]; x[T, 32:33]=0.625 x1[T, 32:33]+0.375 x2[T, 32:33]; x[T, 33:34]=0.75 x1[T, 33:34]+0.25 x2[T, 33:34]; x[T, 34:35]=0.875 x1[T, 34:35]+0.125 x2[T, 34:35]; Of course, other smoothing strategies can also be used. The key is to ensure that within the smoothing range, from low frequency to high frequency, the weight of bone conduction gradually decreases while the weight of air conduction gradually increases.

[0069] Step 3: Input the frequency-domain spliced ​​bone and air conduction audio signal into the pre-trained signal denoising model (the signal denoising model used in this specific embodiment is a neural network model) to predict the clean frequency-domain spliced ​​bone and air conduction audio signal.

[0070] This approach sets the model's prediction target as a clean, frequency-domain spliced ​​audio signal consisting of low-frequency bone conduction signals and high-frequency air conduction audio signals. This avoids predicting low-frequency air conduction components that are completely submerged by noise under high-noise conditions, thus ensuring effective speech denoising in low signal-to-noise ratio scenarios.

[0071] It should be noted that the above examples are only used to help understand this embodiment and do not constitute a limitation on the signal noise reduction process of this embodiment. Any simple modifications based on this technical concept are within the protection scope of this application.

[0072] Based on the first and / or second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to the first and second embodiments described above can be referred to the above description and will not be repeated hereafter. Based on this, the spliced ​​audio signal includes an amplitude spectrum and a phase spectrum. The step of inputting the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal includes: Step E10: Input the amplitude spectrum of the spliced ​​audio signal into the signal denoising model to predict the denoised amplitude spectrum; The spliced ​​audio signal comprises two components in the frequency domain: the amplitude spectrum and the phase spectrum. The amplitude spectrum reflects the energy or amplitude distribution of the signal across its frequency components and is key information characterizing the essential features of the speech signal. The phase spectrum, on the other hand, primarily carries the signal's temporal location information. In speech denoising tasks, the human ear is more sensitive to changes in the amplitude spectrum and less perceptible to subtle deviations in the phase spectrum. Therefore, neural network models typically use the amplitude spectrum as the primary processing target.

[0073] Specifically, the generated spliced ​​audio signal is transformed to the frequency domain using a Fourier transform, and its amplitude spectrum is extracted as input features and fed into a pre-trained signal denoising model. This model is based on a deep learning architecture and is trained on a large amount of clean and noisy speech paired data, enabling it to learn the complex mapping relationship from noisy amplitude spectrum to clean amplitude spectrum. After receiving the amplitude spectrum of the spliced ​​audio signal, the model performs forward propagation calculations through its network structure (such as convolutional neural network, recurrent neural network, or Transformer architecture), progressively extracting high-level feature representations in multiple network layers and suppressing noise-related components. Finally, it predicts and outputs the denoised amplitude spectrum. This denoised amplitude spectrum retains the high signal-to-noise ratio of the bone conduction signal in the low-frequency band while maintaining the speech harmonic details of the air conduction audio signal in the high-frequency band, thus possessing a high-quality spectral structure.

[0074] Step E20: The amplitude spectrum after noise reduction is synthesized with the phase spectrum of the spliced ​​audio signal, and the noise-reduced target audio signal is output after time-domain conversion.

[0075] After obtaining the denoised amplitude spectrum, since the model prediction process only processes the amplitude spectrum and does not change the phase information, it is necessary to combine the denoised amplitude spectrum with the phase spectrum of the original spliced ​​audio signal to reconstruct the complete frequency domain complex spectrum. The phase spectrum retains the phase information of the original spliced ​​audio signal, including the signal's start time and waveform shape in the time domain, and together with the denoised amplitude spectrum, it constitutes a complete frequency domain representation.

[0076] The synthesized complex spectrum is subjected to an inverse Fourier transform, converting it from the frequency domain back to the time domain to obtain a denoised audio signal in time-series form. This time-domain signal retains the noise resistance characteristics contributed by the bone conduction signal in the low-frequency range, allowing the fundamental frequency and low-frequency harmonics of speech to be clearly presented in noisy environments. It also retains the naturalness and detail richness of speech contributed by the air conduction audio signal in the high-frequency range. At the same time, the residual noise is further suppressed by the deep processing of the denoising model. The final output target audio signal has excellent qualities of high clarity, high naturalness and low noise interference in overall auditory perception.

[0077] In one possible implementation, the step of inputting the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal includes: Step F10: Divide the spliced ​​audio signal into multiple sub-band signals; Subband splitting refers to dividing the spliced ​​audio signal into multiple subband intervals with different frequency ranges in the frequency domain. Each subband signal corresponds to a spectral component within a specific frequency band interval.

[0078] Specifically, subband splitting can be achieved through filter banks. For example, a uniform filter bank can be used to divide the spectrum of the spliced ​​audio signal into multiple subbands at equal frequency intervals, with each subband having the same bandwidth. Alternatively, a non-uniform filter bank that conforms to the characteristics of human hearing can be used, such as dividing the frequency bands according to the Buck scale or the Mel scale, so that the subbands in the low-frequency region are divided more densely and the subbands in the high-frequency region are divided more sparsely, in order to match the frequency resolution characteristics of human ears, which are more sensitive to low-frequency signals.

[0079] Each sub-band signal after splitting independently contains spectral information within its corresponding frequency band. Each sub-band signal can be input into the subsequent noise reduction process in parallel or serially, enabling the model to perform differentiated noise reduction processing for the signal characteristics of different frequency bands.

[0080] Step F20: Input each of the sub-band signals into the signal denoising model for denoising processing to obtain the denoised sub-band signals corresponding to each of the sub-band signals; Signal denoising models can employ network structures optimized for subband signal processing. For example, each subband can have its own independent lightweight subnetwork, or a single multi-input multi-output network can be used to process all subband signals simultaneously. After each subband signal is input into the denoising model, the model performs independent denoising mapping on each subband signal, learning the frequency domain transformation relationship from noisy subbands to clean subbands.

[0081] Specifically, since different subbands carry different information sources and noise characteristics in the spliced ​​audio signal, for example, low-frequency subbands mainly originate from bone conduction signals and have a high signal-to-noise ratio, while mid-to-high frequency subbands mainly originate from air conduction audio signals and may contain different levels of noise, the signal denoising model can adaptively adjust the denoising intensity when processing each subband. For low-frequency subbands originating from bone conduction, weaker or even zero denoising processing is applied to maintain their original high signal-to-noise ratio characteristics, while for mid-to-high frequency subbands originating from air conduction, the corresponding denoising intensity is applied according to their noise level, thereby maximizing the preservation of speech structure features while suppressing noise.

[0082] After processing by the model, each sub-band outputs a noise-reduced sub-band signal, and each noise-reduced sub-band signal has optimized spectral quality within its respective frequency band.

[0083] Step F30: Combine the denoised subband signals to obtain the denoised target audio signal.

[0084] Synthesis processing involves recombining the various denoised subband signals in the frequency domain to restore a complete full-band spectrum signal.

[0085] Specifically, following the frequency band division scheme used during sub-band splitting, the corresponding spectral components of each denoised sub-band signal are retained within their respective frequency bands. The spectral components of each sub-band are then arranged and spliced ​​in ascending order of frequency to form a fused spectrum covering the entire frequency band. Next, an inverse Fourier transform is performed on this fused spectrum to convert it back from the frequency domain to the time domain, yielding the denoised target audio signal. If a filter bank was used for sub-band splitting in step F10, then a corresponding synthesis filter bank can be used in step F30 to achieve the comprehensive reconstruction of the sub-band signals.

[0086] Through the above subband splitting, subband-by-subband denoising and subband synthesis processing, the signal denoising model can perform fine-grained differential denoising for different noise characteristics of different frequency bands in the spliced ​​audio signal. This not only protects the original high signal-to-noise ratio of the low-frequency bone conduction component, but also effectively suppresses the residual noise that may exist in the high-frequency components of the air conduction, thereby further improving the overall quality of the target audio signal after denoising.

[0087] In one possible implementation, the step of splitting the spliced ​​audio signal into multiple sub-band signals includes: Step G10: Determine the boundary frequencies of multiple critical frequency bands based on the characteristics of human hearing, wherein the bandwidth of the critical frequency band increases with the increase of the center frequency. The characteristics of human hearing refer to the fact that the human ear's perception of sound frequencies is not linearly distributed, but rather possesses a non-linear frequency resolution capability. This means that the ear has higher resolution in the low-frequency region, capable of distinguishing narrow frequency differences, while its resolution is lower in the high-frequency region, and its sensitivity to frequency differences decreases. The critical band is a fundamental concept in auditory psychology, describing a frequency analysis filter bank on the basilar membrane of the human ear corresponding to a specific frequency range. Frequency components within each critical band interact with each other in auditory perception, producing masking effects and other phenomena.

[0088] Specifically, the boundary frequencies of the critical band can be determined using the Buck scale or the equivalent rectangular bandwidth scale. Under the Buck scale, the bandwidth of the critical band increases approximately linearly with increasing center frequency. For example, in the low-frequency region (around 200Hz), the critical band bandwidth is approximately 100Hz; in the mid-frequency region (around 1000Hz), it is approximately 160Hz; and in the high-frequency region (around 4000Hz), it is approximately 700Hz or higher. The equivalent rectangular bandwidth scale, based on fitting data from a large amount of psychoacoustic experiments, also reflects the increasing bandwidth of the critical band with increasing center frequency and is closer to the actual filtering characteristics of the human ear. Using these scales, a set of critical bands covering the entire audible frequency range can be predetermined. Each critical band is defined by a start frequency and a cutoff frequency, and the critical bands are connected end-to-end to cover the entire frequency band.

[0089] Step G20: According to the boundary frequencies of each critical frequency band, the spliced ​​audio signal is split into multiple sub-band signals that correspond one-to-one with each critical frequency band.

[0090] After determining the boundary frequencies of multiple critical frequency bands, this step divides the spectrum of the spliced ​​audio signal according to these boundary frequencies. Each critical frequency band corresponds to a sub-band signal, and the number of sub-band signals is the same as the number of critical frequency bands. The frequency range of each sub-band signal is consistent with its corresponding critical frequency band.

[0091] Specifically, for the spectrum of the spliced ​​audio signal after Fourier transform, based on the start and cutoff frequencies of each critical band, the frequency points falling within the range of each critical band are extracted to form the spectral components of the sub-band signal corresponding to that critical band. The frequency ranges covered by each sub-band signal are different and do not overlap, arranged sequentially from low frequency to high frequency. All sub-band signals together constitute the complete spectral representation of the spliced ​​audio signal.

[0092] Using the sub-band splitting method based on the characteristics of human hearing, the spliced ​​audio signal is divided into multiple sub-bands that match the laws of auditory perception. Since the human ear has varying sensitivities to noise and distortion in different frequency bands, the subsequent noise reduction model can apply differentiated processing based on the importance of each frequency band in auditory perception when performing independent noise reduction on each sub-band. This makes the noise reduction process more consistent with the characteristics of human auditory perception, thereby maximizing subjective listening quality with limited model complexity and further improving the auditory comfort and naturalness of the target audio signal after noise reduction.

[0093] Furthermore, this application also proposes an audio device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the bone conduction splicing noise reduction method described above. The audio device provided in this application, employing the bone-air conduction splicing noise reduction method described in the above embodiments, can solve the technical problem of how to improve signal noise reduction performance. Compared with the prior art, the beneficial effects of the audio device provided in this application are the same as those of the bone-air conduction splicing noise reduction method provided in the above embodiments, and other technical features of this audio device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0094] refer to Figure 5 It illustrates a structural schematic diagram of an audio device suitable for implementing the embodiments of this application. The audio device in the embodiments of this application may also include, but is not limited to, smart headphones, AR glasses, VR glasses, car speakers, smart speakers, etc. Figure 5The audio device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of this application.

[0095] like Figure 5 As shown, the audio device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the audio device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the system may be connected to the I / O interface 1006 to: input devices 1007, such as touchscreens, touchpads, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008, such as liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003, such as magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows audio devices to communicate wirelessly or wiredly with other devices to exchange data. While audio devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0096] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0097] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0098] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0099] In addition, to achieve the above objectives, this application also provides a readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the bone air conduction splicing noise reduction method in the above embodiments.

[0100] The computer-readable storage medium provided in this application embodiment may be, for example, a USB flash drive, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0101] The aforementioned computer-readable storage medium may be included in an audio device or may exist independently without being assembled into an audio device.

[0102] The aforementioned computer-readable storage medium carries one or more programs that, when executed by an audio device, cause the audio device to perform the process steps of any embodiment of the bone conduction splicing noise reduction method.

[0103] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0104] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0105] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0106] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described bone-air conduction stitching noise reduction method, thereby solving the technical problem of how to improve signal noise reduction effect. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the bone-air conduction stitching noise reduction method provided in the above embodiments, and will not be repeated here.

[0107] Furthermore, embodiments of this application also propose a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described bone conduction splicing and noise reduction method.

[0108] The specific implementation of the computer program product in this application is basically the same as the various embodiments of the above-mentioned bone air conduction splicing noise reduction method, and will not be repeated here.

[0109] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0110] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases, the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software sensor. This computer software sensor is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause an audio device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0112] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for noise reduction through bone air conduction splicing, characterized in that, The bone air conduction splicing noise reduction method includes the following steps: Acquire air conduction audio signals, bone conduction audio signals, and a pre-trained signal denoising model; The first component signal belonging to the first frequency band is extracted from the bone conduction audio signal, and the second component signal belonging to the second frequency band is extracted from the air conduction audio signal, wherein the first frequency band and the second frequency band belong to different frequency band intervals in the frequency domain, and the upper limit of the first frequency band is less than the upper limit of the second frequency band. The first component signal and the second component signal are frequency domain merged to generate a spliced ​​audio signal; The spliced ​​audio signal is input into the signal denoising model to obtain the denoised target audio signal.

2. The bone air conduction splicing noise reduction method as described in claim 1, characterized in that, The step of frequency-domain merging of the first component signal and the second component signal to generate a spliced ​​audio signal includes: Within the transition frequency band between the first frequency band and the second frequency band, the frequency point values ​​of the first component signal and the second component signal are weighted and merged according to a preset weight, and the merging result is used as the value of the spliced ​​audio signal at the corresponding frequency point. Outside the transition frequency band between the first frequency band and the second frequency band, the frequency point value of the first component signal in the first frequency band and the frequency point value of the second component signal in the second frequency band are respectively used as the value of the spliced ​​audio signal at the corresponding frequency point.

3. The bone air conduction splicing noise reduction method as described in claim 2, characterized in that, The step of weighting and merging the frequency values ​​of the first component signal and the second component signal according to a preset weight includes: The weights corresponding to the first component signal and the second component signal are determined respectively, wherein, from the low-frequency end to the high-frequency end of the transition frequency band, the weight of the first component signal decreases and the weight of the second component signal increases. The frequency values ​​of the first component signal and the second component signal are weighted and combined according to the determined weights.

4. The bone air conduction splicing noise reduction method as described in claim 3, characterized in that, The method further includes the step of determining the weights corresponding to the first component signal and the second component signal respectively: Calculate the local signal-to-noise ratio of the air-conducted audio signal within the transition frequency band; The initial weights of the first component signal and the second component signal are determined based on the local signal-to-noise ratio, wherein the higher the local signal-to-noise ratio, the higher the initial weight of the second component signal and the lower the initial weight of the first component signal. Based on the initial weights, the weights corresponding to the first component signal are increased sequentially according to a preset step size, and the weights corresponding to the second component signal are decreased sequentially.

5. The bone air conduction splicing noise reduction method as described in claim 2, characterized in that, After the steps of acquiring air conduction audio signals, bone conduction audio signals, and a pre-trained signal denoising model, the method further includes: Determine the global signal-to-noise ratio of the air-conducted audio signal; When the global signal-to-noise ratio is lower than a preset signal-to-noise ratio threshold, the upper limit frequency of the first frequency band is extended to expand the bandwidth of the transition band. When the global signal-to-noise ratio is higher than the preset signal-to-noise ratio threshold, the upper limit frequency of the first frequency band is reduced to decrease the bandwidth of the transition frequency band.

6. The bone air conduction splicing noise reduction method as described in claim 1, characterized in that, The spliced ​​audio signal includes an amplitude spectrum and a phase spectrum. The step of inputting the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal includes: The amplitude spectrum of the spliced ​​audio signal is input into the signal denoising model to predict the denoised amplitude spectrum; The amplitude spectrum after noise reduction is synthesized with the phase spectrum of the spliced ​​audio signal, and the noise-reduced target audio signal is output after time-domain transformation.

7. The bone air conduction splicing noise reduction method as described in claim 1, characterized in that, The step of inputting the spliced ​​audio signal into the signal denoising model to obtain the denoised target audio signal includes: The spliced ​​audio signal is split into multiple sub-band signals; Each of the sub-band signals is input into the signal denoising model for denoising processing to obtain the denoised sub-band signal corresponding to each of the sub-band signals. The denoised sub-band signals are synthesized to obtain the denoised target audio signal.

8. The bone air conduction splicing noise reduction method as described in claim 7, characterized in that, The step of splitting the spliced ​​audio signal into multiple sub-band signals includes: The boundary frequencies of multiple critical frequency bands are determined based on the characteristics of human hearing, wherein the bandwidth of the critical frequency band increases with the increase of the center frequency. According to the boundary frequencies of each of the critical frequency bands, the spliced ​​audio signal is divided into multiple sub-band signals that correspond one-to-one with each of the critical frequency bands.

9. An audio device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the bone air conduction splicing noise reduction method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the bone air conduction splicing noise reduction method as described in any one of claims 1 to 8.

Images