Multi-scene adaptive noise reduction method and system for audio terminal equipment

Abstract

The invention relates to the technical field of audio noise reduction, in particular to a multi-scene adaptive noise reduction method and system for audio terminal equipment. The method comprises the following steps: identifying a real-time environmental sound signal; environment noise source identification is carried out on the real-time environment sound signal, noise distribution perception modeling is carried out, and an environment noise distribution field is constructed; performing current scene noise change analysis on the environment noise distribution field, performing dynamic noise distribution rendering, and constructing a dynamic noise scene model; performing hedging suppression parameter calculation based on the dynamic noise scene model, performing adaptive noise suppression processing, and constructing an adaptive noise suppression strategy; acquiring an audio input signal of a user; and performing local audio gain processing on an audio input signal of a user, and performing multi-band decomposition to generate an audio frequency parameter of each band. The stability and the accuracy of the noise reduction effect are improved, so that the practicability and the adaptability of the audio terminal equipment are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of audio noise reduction, and particularly to a multi-scenario adaptive noise reduction method and system for audio terminal devices. Background Art

[0002] As an important tool for daily communication and entertainment, audio terminal devices are widely used in fields such as smart phones, earphones, and smart speakers, and undertake key tasks such as receiving, processing, and outputting audio signals. During use, the sound quality and call quality of audio terminal devices directly affect the user experience. With the continuous progress of speech recognition, voice call, and audio playback technologies, users' requirements for audio quality are getting higher and higher. Especially in noisy environments, the interference of background noise has become a major challenge in audio processing.

[0003] Traditional noise reduction technologies mainly rely on fixed algorithms for noise suppression. However, in changing application scenarios, these methods are often difficult to cope with complex noise environments. When using earphones on a crowded street, traditional noise reduction algorithms cannot effectively distinguish noise from speech signals, resulting in unclear sound quality; while in a quiet environment, excessive noise reduction will cause speech signal distortion and affect call quality. Most of the existing noise reduction technologies are optimized for a single environment, lacking flexibility and adaptability, and unable to automatically adjust to the best noise reduction state in different scenarios. Therefore, in order to improve the audio processing ability of audio terminal devices in changing scenarios, a multi-scenario adaptive noise reduction method that can adapt to environmental changes in real time is needed. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a multi-scenario adaptive noise reduction method and system for audio terminal devices to solve at least one of the above technical problems.

[0005] To achieve the above object, the present invention provides a multi-scenario adaptive noise reduction method for an audio terminal device, including the following steps: Step S1: Identify the real-time environmental sound signal; identify the environmental noise source of the real-time environmental sound signal, and perform noise distribution perception modeling to construct an environmental noise distribution field; Step S2: Analyze the current scene noise change of the environmental noise distribution field, and perform dynamic noise distribution rendering to construct a dynamic noise scene model; Step S3: Calculate the hedging suppression parameter based on the dynamic noise scene model, and perform adaptive noise suppression processing to construct an adaptive noise suppression strategy; Step S4: Obtain the user's audio input signal; perform local audio gain processing on the user's audio input signal, and perform multi-band decomposition to generate audio frequency parameters for each band; Step S5: Fine-tune the noise resonance frequency of the dynamic noise scenario model according to the audio frequency parameters of each band, and perform instant distortion compensation processing to construct a globally reconstructed and optimized audio signal; Step S6: Perform multi-scenario transfer learning on the globally reconstructed and optimized audio signal, and perform dynamic iterative optimization according to the adaptive noise suppression strategy to construct an intelligent multi-scenario noise reduction engine.

[0006] Through the recognition of environmental sound signals, the present invention can accurately detect and classify noise sources in the environment. By identifying environmental noise sources, the system can perceive the distribution of different noise types, thereby providing an important basis for subsequent noise suppression and scene modeling. The perception and modeling of noise distribution can help the system establish a detailed noise distribution field, accurately reflecting the spatial distribution characteristics of the current environmental noise, and ensuring that the noise reduction process is more targeted and precise. Through the dynamic analysis of noise changes, the system can real-time perceive the changing trend of noise, such as the change in noise intensity or the switching of noise types. Based on these changes, dynamically rendering the noise scene model can achieve accurate modeling of noise in different scenarios, thereby providing real-time and highly adaptable basic data for subsequent noise suppression and optimization. The construction of the dynamic noise scene model helps the system better cope with complex and variable environmental noise, improving the real-time and accuracy of the noise reduction effect. By calculating the hedge suppression parameters based on the dynamic noise scene model, the best noise suppression parameters can be accurately evaluated and calculated, and adaptive noise suppression processing can be achieved. The core of this step is to make the noise reduction strategy automatically adjust according to real-time noise changes to achieve the optimal noise reduction effect. The construction of the adaptive noise suppression strategy means that the system can automatically adjust according to the real-time changes of environmental noise, continuously optimizing the processing effect, and avoiding problems of over-suppression or under-suppression caused by fixed noise reduction parameters. By performing local audio gain processing on the user audio input signal, the quality of the audio signal is optimized in different frequency bands, improving its clarity and listening experience. The multi-band decomposition technology helps to disassemble the audio signal into multiple frequency bands, enabling each frequency band to be optimized individually, avoiding negative impacts on the entire audio signal. In addition, generating the frequency parameters of each band helps to accurately perform subsequent noise suppression processing and frequency adjustment for each frequency band. By finely adjusting the frequency parameters of each band, the resonance frequency of noise can be effectively suppressed, avoiding interference and distortion caused by similar noise frequencies. In addition, the instant distortion compensation processing effectively corrects the audio distortion that occurs during the noise reduction process, ensuring the naturalness and clarity of the audio output. The global reconstruction optimizes the construction of the audio signal, ensuring that the audio quality of all frequency bands is improved, and at the same time optimizing the overall quality of the audio signal. Through multi-scenario transfer learning, the system can learn and optimize the noise suppression strategy from audio data in different environments and scenarios, thereby improving the adaptive ability in different environments. The dynamic iterative optimization enables the noise reduction strategy to continuously self-optimize according to different scenarios during actual use, improving the stability and accuracy of the noise reduction effect. The construction of the intelligent multi-scenario noise reduction engine enables the system to cope with more complex noise environments, greatly enhancing the practicality and adaptability of audio terminal devices and improving the user experience.

[0007] In this specification, a multi-scenario adaptive noise reduction system for an audio terminal device is provided, which is used to execute the multi-scenario adaptive noise reduction method of the audio terminal device as described above, including: A noise perception module, configured to identify real-time environmental sound signals; identify environmental noise sources for the real-time environmental sound signals, and perform noise distribution perception modeling to construct an environmental noise distribution field; A noise distribution rendering module, configured to perform current scene noise change analysis on the environmental noise distribution field, and perform dynamic noise distribution rendering to construct a dynamic noise scene model; A noise suppression module, configured to calculate hedge suppression parameters based on the dynamic noise scene model, and perform adaptive noise suppression processing to construct an adaptive noise suppression strategy; A local audio gain module, configured to obtain the user's audio input signal; perform local audio gain processing on the user's audio input signal, and perform multi-band decomposition to generate audio frequency parameters for each band; A resonance frequency fine-tuning module, configured to fine-tune the noise resonance frequency of the dynamic noise scene model according to the audio frequency parameters of each band, and perform instant distortion compensation processing to construct a globally reconstructed and optimized audio signal; An intelligent noise reduction optimization module, configured to perform multi-scenario transfer learning on the globally reconstructed and optimized audio signal, and perform dynamic iterative optimization according to the adaptive noise suppression strategy to construct an intelligent multi-scenario noise reduction engine.

[0008] The present invention can comprehensively collect sound signals in the current environment by using multimodal input of microphone array and sensor. This lays the foundation for the subsequent location, classification and processing of noise sources. By analyzing the audio characteristics (such as frequency, intensity, time domain changes, etc.) of different noise sources in the environment, the system can identify the types of noise sources in the environment (such as wind noise, traffic noise, conversation sounds, etc.). Through the application of acoustic models, the device can establish a noise distribution field to help understand the distribution of noise sources in space, which is convenient for subsequent noise suppression and optimization processing. Through dynamic noise analysis, the device can monitor the changes in noise in the environment in real time and automatically adapt to noise changes in different scenarios. During commuting, the noise source is constantly changing, and noise sources such as vehicles passing by and road construction are constantly switching. The system can accurately identify these changes and adjust the noise reduction strategy. By modeling the time-varying noise source, the device can construct a dynamic noise scene model to adapt it to the instantaneous changing environmental noise. This not only enhances the accuracy of the noise reduction algorithm, but also provides accurate noise source location and noise type data for the subsequent adaptive noise reduction algorithm. By calculating the dynamic noise scene model, the device can calculate effective noise countermeasure parameters according to the intensity, location and propagation path of the noise source. This countermeasure method reduces the interference of noise on the audio signal by generating a sound waveform that counteracts the noise source. The device can automatically adjust the noise suppression strategy according to the changes and characteristics of the noise in the current environment. Compared with the traditional fixed parameter noise reduction method, adaptive noise suppression can make precise adjustments according to the noise intensity, spectrum and scene requirements to ensure the best noise reduction effect in different environments. As the environmental noise changes, the system can dynamically adapt to these changes, not only reducing common noise (such as traffic noise, wind noise, etc.), but also dealing with transient noise (such as sudden noise or human voice interference). By processing the local gain of the user's audio signal, the clarity and quality of the user's voice can be highlighted to avoid voice distortion caused by excessive noise suppression. At the same time, multi-band decomposition technology enables the device to accurately process audio signals in different frequency bands, thereby effectively distinguishing between voice and noise. After the device generates frequency parameters for each audio band, it can more accurately identify and suppress noise according to the characteristics of different frequencies. There are different processing strategies for low-frequency noise and high-frequency noise. This frequency band decomposition allows the device to flexibly cope with diverse noise environments. Based on the frequency parameters of each audio band, the system can find and adjust the resonant frequency close to the frequency of the noise source. This step can optimize the spectrum of the audio signal so that the noise in the resonant frequency band is effectively suppressed while retaining the naturalness of the speech. By compensating the distortion of the audio signal, the system can correct the sound quality loss caused by the noise reduction process. This real-time compensation ensures that the device maintains the clarity and naturalness of the audio during the noise suppression process.By optimizing the synthesis of noise for each band, the device can output a globally reconstructed audio signal, ensuring that users receive high-quality and clear sound, whether in a quiet environment or a noisy scene. The device can learn effective noise reduction strategies from different noise scenarios through transfer learning technology and apply them to new scenarios. Transfer learning enables the system to be adaptive from the initial stage, reducing training time and improving noise reduction efficiency. According to environmental changes and user needs, the device will continuously self-learn and optimize to ensure optimal noise reduction effects in multiple scenarios. When the user enters a noisy public transportation vehicle or a meeting, the device can automatically identify the scenario and adjust the noise reduction strategy. Through continuous iteration and optimization, the system can adjust the noise reduction parameters in real time in different scenarios, avoiding the static setting of traditional noise reduction methods, enabling precise noise suppression in each environment, and enhancing the user's audio experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 is a schematic flowchart of the steps of a multi-scenario adaptive noise reduction method for an audio terminal device of the present invention; Figure 2 is a schematic flowchart of the detailed implementation steps of step S1; Figure 3 is a schematic flowchart of the detailed implementation steps of step S2; Figure 4 is a schematic flowchart of the detailed implementation steps of step S3. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0010] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0011] The embodiments of the present application provide a multi-scenario adaptive noise reduction method and system for an audio terminal device. The execution subjects of the multi-scenario adaptive noise reduction method and system for the audio terminal device include, but are not limited to, mechanical devices, data processing platforms, cloud server nodes, network upload devices, etc. that carry this system, which can be regarded as general computing nodes of the present application. The data processing platform includes, but is not limited to, at least one of an audio image management system, an information management system, and a cloud data management system.

[0012] Please refer to Figures 1 to 4 , the present invention provides a multi-scenario adaptive noise reduction method for an audio terminal device. The multi-scenario adaptive noise reduction method for the audio terminal device includes the following steps: Step S1: Identify the real-time environmental sound signal; identify the environmental noise source of the real-time environmental sound signal, and perform noise distribution perception modeling to construct an environmental noise distribution field; Step S2: Analyze the current scene noise variation of the environmental noise distribution field, perform dynamic noise distribution rendering, and construct a dynamic noise scene model; Step S3: Calculate the hedging suppression parameters based on the dynamic noise scene model, perform adaptive noise suppression processing, and construct an adaptive noise suppression strategy; Step S4: Obtain the user's audio input signal; perform local audio gain processing on the user's audio input signal and conduct multi-band decomposition to generate audio frequency parameters for each band; Step S5: Fine-tune the noise resonance frequency of the dynamic noise scene model according to the audio frequency parameters of each band, and perform instant distortion compensation processing to construct a globally reconstructed and optimized audio signal; Step S6: Perform multi-scene transfer learning on the globally reconstructed and optimized audio signal, and conduct dynamic iterative optimization according to the adaptive noise suppression strategy to construct an intelligent multi-scene noise reduction engine.

[0013] Through the recognition of environmental sound signals, the present invention can accurately detect and classify noise sources in the environment. Through the recognition of environmental noise sources, the system can perceive the distribution of different noise types, thus providing an important basis for subsequent noise suppression and scene modeling. The perception and modeling of noise distribution can help the system establish a detailed noise distribution field, accurately reflecting the spatial distribution characteristics of the current environmental noise, and ensuring that the noise reduction process is more targeted and precise. Through the dynamic analysis of noise changes, the system can real-time perceive the changing trend of noise, such as the change in noise intensity or the switching of noise types. Based on these changes, dynamically rendering the noise scene model can achieve accurate modeling of noise in different scenarios, thus providing real-time and highly adaptable basic data for subsequent noise suppression and optimization. The construction of the dynamic noise scene model helps the system better cope with complex and changing environmental noise, improving the real-time and accuracy of the noise reduction effect. By calculating the hedging suppression parameters based on the dynamic noise scene model, the optimal noise suppression parameters can be accurately evaluated and calculated, and adaptive noise suppression processing can be achieved. The core of this step is to make the noise reduction strategy automatically adjust according to real-time noise changes to achieve the optimal noise reduction effect. The construction of the adaptive noise suppression strategy means that the system can automatically adjust according to the real-time changes of environmental noise, continuously optimizing the processing effect, and avoiding problems of over-suppression or under-suppression caused by fixed noise reduction parameters. By performing local audio gain processing on the user audio input signal, the quality of the audio signal is optimized in different frequency bands, improving its clarity and listening experience. The multi-band decomposition technique helps to disassemble the audio signal into multiple frequency bands, enabling each frequency band to be optimized separately and avoiding negative impacts on the entire audio signal. In addition, generating the frequency parameters of each band helps to precisely perform subsequent noise suppression processing and frequency adjustment for each frequency band. By finely adjusting the frequency parameters of each band, the resonance frequency of the noise can be effectively suppressed, avoiding interference and distortion caused by similar noise frequencies. In addition, the instant distortion compensation processing effectively corrects the audio distortion that occurs during the noise reduction process, ensuring the naturalness and clarity of the audio output. The global reconstruction optimizes the construction of the audio signal, ensuring that the audio quality of all frequency bands is improved, and at the same time optimizing the overall quality of the audio signal. Through multi-scenario transfer learning, the system can learn and optimize the noise suppression strategy from audio data in different environments and scenarios, thus improving the adaptive ability in different environments. The dynamic iterative optimization enables the noise reduction strategy to continuously self-optimize according to different scenarios during actual use, improving the stability and accuracy of the noise reduction effect. The construction of the intelligent multi-scenario noise reduction engine enables the system to cope with more complex noise environments, greatly enhancing the practicality and adaptability of audio terminal devices and improving the user experience.

[0014] In an embodiment of the present invention, refer to Figure 1, which is a schematic diagram of the step flow of a multi-scenario adaptive noise reduction method for an audio terminal device of the present invention. In this example, the steps of the multi-scenario adaptive noise reduction method for the audio terminal device include: Step S1: Identify the real-time environmental sound signal; identify the environmental noise source of the real-time environmental sound signal, and perform noise distribution perception modeling to construct an environmental noise distribution field; In this embodiment, a multi-channel microphone array is used to collect real-time sound signals in order to capture various sounds in the environment. The selected microphone array should have at least four channels to improve spatial resolution and sound localization ability. A suitable sampling rate (such as 16 kHz or 44.1 kHz) is set to ensure the clarity and accuracy of the audio signal. Real-time recording is performed using the pyaudio library in Python or MATLAB. Preprocessing is performed on the collected audio signal, including denoising, removing the DC component, and normalization. Common denoising techniques include spectral subtraction and wavelet denoising, and the librosa library is used for audio signal processing. Denoising parameters, such as the wavelet decomposition level and threshold, are set to ensure a clear signal for subsequent analysis. A machine learning or deep learning model is used to identify the noise source of the real-time environmental sound signal. Common models include convolutional neural networks (CNNs) and recurrent neural networks (RNNs). Open-source libraries (such as TensorFlow or PyTorch) are used to build and train the model. A labeled audio dataset (such as UrbanSound or ESC-50) is used for model training to improve the model's ability to identify different noise sources. The preprocessed audio signal is framed (one frame every 20 milliseconds), and features are extracted for each frame. Common features include Mel-frequency cepstral coefficients (MFCCs), spectrograms, and zero-crossing rates, etc. The extracted features are input into the trained noise source identification model to identify the type of noise source in the environment in real time (such as traffic noise, human voices, mechanical noise, etc.). Based on the identified noise source and its corresponding feature information, an environmental noise distribution model is constructed. Spatial statistical methods (such as Kriging interpolation) are used for modeling to estimate the noise distribution throughout the environment. Parameters of the noise distribution model are set, such as the intensity, location, and propagation characteristics of the noise source. Adjustments are made according to the actual environment to enhance the accuracy of the model. Data visualization tools (such as Matplotlib or Plotly) are used to visualize the constructed noise distribution field, showing the noise intensity distribution in different regions. A color mapping is set to help users intuitively understand the noise distribution. A heat map or three-dimensional visualization map is generated to clearly identify the noise source and its influence range. A dynamic update mechanism is set so that the noise distribution field can reflect environmental changes in real time. New audio signals are obtained regularly, the noise source is re-identified, and the model parameters are updated. The update frequency is set, for example, to update once a minute to ensure that the model always reflects the current environmental noise situation. The construction results of the noise distribution field are recorded in a database, including the noise source location, intensity, and distribution model parameters, etc., for subsequent analysis and optimization. Based on the constructed noise distribution field, the impact of noise on the environment is further analyzed to provide a basis for noise control and management.

[0015] Step S2: Analyze the current scene noise change of the environmental noise distribution field, perform dynamic noise distribution rendering, and construct a dynamic noise scene model; In this embodiment, after the environmental noise distribution field is constructed, the change of environmental noise is continuously monitored. A multi-channel microphone array is used to periodically collect audio signals, and the sampling frequency is set to 44.1 kHz to ensure high-quality audio data. Define a time window (for example, every 30 seconds), and collect audio data within this window for subsequent analysis. Preprocess the audio signals within each time window, including denoising and framing. Use the short-time Fourier transform (STFT) to extract spectral features, set the frame length to 20 milliseconds, and the overlap rate to 50%. The extracted features include Mel-frequency cepstral coefficients (MFCC), energy, and zero-crossing rate, etc., and are processed using the librosa library in Python. Use statistical methods (such as mean and variance) to analyze each feature and compare the noise characteristics of the current time window with those of the previous window. Set a threshold (such as a change amplitude greater than 10%) to determine whether the noise has changed significantly. Record the changing noise sources and their characteristics, including time, intensity, and frequency information, for subsequent modeling. According to the results of the current noise change analysis, construct a dynamic noise distribution rendering model. Use computer graphics technology to visualize the noise distribution in space. Select a suitable rendering tool (such as Unity or OpenGL) to achieve dynamic rendering, ensuring that the noise distribution scene can be updated in real time. Assign colors to different intensities of noise. Set low-intensity noise to green, medium-intensity to yellow, and high-intensity to red. Use a gradient color mapping to visually display the noise intensity. In the rendering, use a grid or a 3D model to represent the environmental structure, and generate a dynamic noise scene in combination with the position and intensity of the noise sources. Set a dynamic update mechanism so that the noise distribution field can reflect the noise changes in the scene in real time. Whenever a noise change is detected, automatically refresh the rendering scene and update the noise intensity and distribution. Use an event-driven architecture to listen for noise change events and trigger corresponding rendering updates. Combine the dynamic noise distribution rendering with the environmental noise distribution field to construct a complete dynamic noise scene model. Ensure that the model can reflect the noise changes in the environment and their impact on the surrounding environment. Use a physics engine (such as PhysX in Unity) to simulate noise propagation, considering factors such as reflection, scattering, and sound absorption to enhance the authenticity of the rendering. Design a user interface that allows users to observe and analyze the noise distribution. Provide real-time data feedback, such as the current noise intensity, change trend, and historical data. Record the user's feedback and usage data for subsequent optimization of the model and rendering effect. Record the results of the dynamic noise distribution field in a database, including noise source location, intensity, change characteristics, and rendering parameters, etc. Ensure that the data structure is clear for subsequent access and analysis. Set the data update frequency, for example, update once per minute, to ensure that the model always reflects the current environmental noise situation.

[0016] Step S3: Calculate the hedging suppression parameters based on the dynamic noise scene model, and perform adaptive noise suppression processing to construct an adaptive noise suppression strategy; In this embodiment, cancellation suppression is a technique for eliminating or reducing noise by adjusting the phase and amplitude of a signal. According to the dynamic noise scenario model, the main noise sources and their frequency characteristics are identified, and the cancellation suppression parameters are calculated. The frequency range of the noise source (such as 20 Hz to 20 kHz) and the target signal frequency are set, and the differences between the noise spectrum and the target spectrum are analyzed. The fast Fourier transform (FFT) is used to perform spectral analysis on the noise signal in the dynamic noise scenario model to extract the main components of the noise. The cancellation suppression parameters, including gain and phase compensation, are calculated. The gain is set as the ratio of the noise amplitude to the target signal amplitude. According to the calculated cancellation suppression parameters, an adaptive filter is designed. Commonly used adaptive filters include the least mean square (LMS) and recursive least squares (RLS) filters. The LMS algorithm is implemented using the scipy library in Python, and the step size (such as 0.01) and filter order (such as 32) of the filter are set to ensure that the filter can effectively track noise changes. The real-time audio signal is frame-processed, with each frame length set to 20 milliseconds and an overlap rate of 50%. The adaptive filter is applied to each frame, and the filter coefficients are dynamically adjusted to eliminate noise. After each processing, the calculated gain and phase compensation are used to adjust the output signal to ensure the improvement of the quality of the target signal. During the noise suppression process, the signal-to-noise ratio (SNR) of the signal is monitored in real time, and the target SNR value (such as 10 dB) is set to ensure that the output signal after being processed by the adaptive filter meets the standard. The SNR value and filter adjustment parameters of each processing are recorded for subsequent analysis and optimization. According to the dynamic noise scenario model and the results of the adaptive noise suppression process, a complete set of adaptive noise suppression strategies is constructed. The strategy should include links such as noise source identification, cancellation suppression parameter calculation, and adaptive filter design. A dynamic update mechanism for the strategy is set. When the ambient noise changes, the cancellation suppression parameters are automatically recalculated and the filter settings are adjusted. The adaptive noise suppression strategy is optimized, such as using an adaptive step size algorithm to dynamically adjust the step size of the filter according to the characteristics of the real-time signal to improve the noise suppression effect. An online learning algorithm is used to update the model parameters to ensure that the strategy can maintain good performance in a changing environment. The adaptive noise suppression strategy is tested under different ambient noise conditions, and the changes in the signal-to-noise ratio before and after noise reduction and user feedback are recorded. The A / B test method is used to compare the effects of different strategies. The implementation results of the strategy and user feedback are recorded in a database to provide a basis for subsequent optimization.

[0017] Step S4: Obtain the audio input signal of the user; perform local audio gain processing on the audio input signal of the user and perform multi-band decomposition to generate audio frequency parameters for each band; In this embodiment, local audio gain processing is performed based on the user's audio input signal. The purpose of local gain is to enhance the audio signal within a specific frequency range to improve the clarity and audibility of the audio. The dynamic range compression (DRC) technique is used to perform gain processing on the signal. The compression ratio (such as 2:1) and threshold (such as -10 dB) are set so that signals exceeding the threshold are compressed. The filter design and application functions in the scipy.signal library are used to adjust the gain of the audio signal. First, a band-pass filter is set, and the frequency range (such as 300 Hz to 3 kHz) is selected to enhance the vocal frequency band. For each time period, the gain value is calculated and applied to the audio signal. The audio after gain processing is saved as a new audio file for subsequent analysis. After gain processing, metrics such as signal-to-noise ratio (SNR) and total harmonic distortion (THD) are used to evaluate the effect of gain processing. The goal is to increase the SNR by at least 5 dB while ensuring that the distortion is controlled within an acceptable range (such as less than 1%). Multi-band decomposition is performed to extract the frequency parameters of the audio signal. Common methods include wavelet transform and filter bank techniques. Here, wavelet transform is used for its advantages in localization in the time and frequency domains. The number of decomposition layers (such as 3 layers) is determined to extract signal features in different frequency ranges. The pywt library (Python's wavelet transform library) is used to perform wavelet decomposition on the audio signal after gain processing. The wavelet basis function (such as Daubechies wavelet) and the number of decomposition layers are set. Wavelet coefficients are extracted in each layer to generate audio signals in different frequency bands. By setting the frequency range (low frequency, middle frequency, and high frequency), the audio signal is divided into multiple bands. The audio signal of each band is saved as a separate audio file, and the frequency parameters of each band (such as center frequency, bandwidth, etc.) are recorded. These parameters are saved in JSON format or a database for subsequent analysis. The center frequency range is set, for example, low frequency (20 Hz to 250 Hz), middle frequency (250 Hz to 2 kHz), and high frequency (2 kHz to 20 kHz), to ensure a comprehensive analysis of the audio signal.

[0018] Step S5: Fine-tune the noise resonance frequency of the dynamic noise scenario model according to the audio frequency parameters of each band, and perform instant distortion compensation processing to construct a globally reconstructed and optimized audio signal; In this embodiment, the noise resonance frequency refers to the frequency within a specific frequency range at which the noise has a significant impact on the system response. In the dynamic noise scenario model, identifying and fine-tuning these resonance frequencies is an important step in optimizing the audio signal quality. Extract the audio frequency parameters of each band from the previous band analysis, including the center frequency, bandwidth, gain, etc. Use digital filters to fine-tune the noise resonance frequencies of the dynamic noise scenario model. Common filters include Butterworth filters and band-pass filters. Select the appropriate filter type to match the target frequency. Set a frequency range (such as ±5% of the center frequency) for fine-tuning to ensure that the audio signal near the resonance frequency is enhanced. Use the scipy.signal library in Python to design and apply the filters. For each band, first calculate its resonance frequency, apply the filter to the audio signal in the dynamic noise scenario model, and update the signal in real time to reflect the effect of fine-tuning. Monitor the characteristics of the fine-tuned audio signal, and use spectrum analysis (such as FFT) to evaluate the effect of resonance frequency fine-tuning. The goal is to increase the signal strength of the target frequency while maintaining the clarity of the signal. Record the signal-to-noise ratio (SNR) of the signal, hoping to increase it by at least 3 dB, and ensure that the distortion rate is controlled within an acceptable range (such as less than 1%). In audio signal processing, distortion will cause the signal quality to decline, especially after gain processing and frequency fine-tuning. Therefore, implementing distortion compensation is a key step in ensuring the audio signal quality. Use non-linear distortion compensation techniques to process the fine-tuned signal. Adopt a method that combines dynamic range compression (DRC) and phase compensation. Dynamic range compression effectively reduces the peak value of the signal, thereby reducing distortion, while phase compensation adjusts the phase of the signal to maintain the natural feeling of the audio. Set the compression ratio (such as 4:1) and threshold (such as -6 dB) to compensate for the signal exceeding the threshold. Use the filters in the scipy.signal library to implement dynamic range compression. First, calculate the signal strength of each time period and apply compression according to the set threshold. At the same time, use phase compensation techniques to adjust the phase of the signal. By calculating the phase difference between the resonance frequency and the target frequency, synthesize the compensated signal with the fine-tuned signal to generate the final optimized audio signal. Integrate the fine-tuned signal with the signal after distortion compensation to generate the globally reconstructed optimized audio signal. By processing each band, ensure that the final output signal is balanced in each frequency range. Use weighted average or mixing techniques to fuse the signals of different bands into a complete audio signal. Set the weight of each band according to its importance and the characteristics of the target signal for adjustment.

[0019] Step S6: Perform multi-scenario transfer learning on the globally reconstructed optimized audio signal, and perform dynamic iterative optimization according to the adaptive noise suppression strategy to construct an intelligent multi-scenario noise reduction engine.

[0020] In this embodiment, multi-scenario audio datasets are collected and constructed. These datasets should contain audio samples under various environmental noise conditions, such as indoor, outdoor, noisy environments, and quiet environments. Each sample is preferably labeled with information such as noise type, intensity, and signal quality. Set the number of samples. It is recommended that each scenario contain at least 500 audio samples to ensure the generalization ability of the model. Use open-source datasets (such as Urban Sound or ESC-50) as a basis and expand them. Extract features from each audio sample using features such as Mel Frequency Cepstral Coefficients (MFCC), spectrograms, and zero-crossing rates. Set the frame length to 25 milliseconds and the overlap rate to 50%. Use the librosa library in Python for feature extraction and store the extracted features and corresponding labels (such as noise type, scenario, etc.) as a feature matrix for subsequent training. Select a suitable deep learning model for multi-scenario transfer learning. Commonly used models include Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs). Select a suitable model according to the characteristics of the dataset. Use TensorFlow or PyTorch to build the model architecture, set the number of layers (such as convolutional layers, pooling layers, and fully connected layers), and select a suitable activation function (such as ReLU). Use the method of transfer learning to fine-tune from a pre-trained model (a model trained on a similar task). Freeze the weights of some convolutional layers and only train the last few layers to adapt to the new scenario data. Set the learning rate (such as 0.001) and batch size (such as 32), and perform model training. Use cross-entropy as the loss function and optimize it using the Adam optimizer. At the end of each training epoch, evaluate the model performance using the validation set and record the accuracy and loss values. Set a threshold (such as an accuracy of 0.85) as the performance standard of the model. Adopt early stopping to prevent overfitting and stop training when the validation loss does not improve for 10 consecutive epochs. Combine the adaptive noise suppression strategy with the multi-scenario transfer learning model to build an intelligent multi-scenario noise reduction engine. The engine should be able to dynamically adjust parameters according to the real-time environmental noise. Set up a feedback mechanism to monitor the environmental noise and user audio input in real-time and automatically adjust the model parameters to optimize the noise reduction effect. During the real-time audio processing, use an online learning algorithm (such as Stochastic Gradient Descent) to dynamically update the model parameters. Set a learning rate decay strategy to adapt to the noise characteristics in different scenarios. Regularly evaluate the model performance, set an update period (such as every minute), and adjust the weights and biases of the model according to the real-time feedback within each period. Verify the effect of the noise reduction engine through multi-scenario tests and record metrics such as Signal-to-Noise Ratio (SNR) and Total Harmonic Distortion (THD) to ensure that the noise reduction effect meets the standards in different scenarios. It is expected that the SNR is increased by at least 5 dB and the THD is controlled within 1%. According to the test results, adjust the adaptive noise suppression strategy, such as re-setting the gain and filter parameters, to improve the noise reduction effect. Deploy the trained intelligent multi-scenario noise reduction engine to practical applications, such as mobile devices or desktop applications.Ensure that the system can process audio signals in real time and apply noise reduction strategies. Set up a user interface that allows users to select different noise reduction modes (such as indoor, outdoor, conference, etc.) and view the noise reduction effect in real time.

[0021] In this embodiment, refer to Figure 2 , which is a schematic diagram of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Step S11: Identify real-time environmental sound signals based on the multi-sensor array of the audio terminal device; Step S12: Identify environmental noise sources for the real-time environmental sound signals and mark each environmental noise source node; Step S13: Perform three-dimensional spatial positioning on each environmental noise source node to obtain the spatial position coordinates of each noise source; Step S14: Analyze the environmental noise distribution for the spatial position coordinates of each noise source to obtain the environmental noise spatial distribution characteristics; Step S15: Perform noise distribution perception modeling based on the environmental noise spatial distribution characteristics and construct an environmental noise distribution field.

[0022] In this embodiment, an audio terminal device with multi-channel input is selected, such as an array microphone. These devices should have high sensitivity and be able to capture sounds in a wide frequency range (e.g., 20 Hz to 20 kHz). The signal-to-noise ratio (SNR) of the preferred device should be greater than 60 dB to ensure that weak sound signals in the environment can be effectively identified. Multiple microphones are arranged in a linear array or a two-dimensional array. A linear array is suitable for sound source localization in one-dimensional space, while a two-dimensional array can improve the ability to capture surrounding sounds. The microphone spacing should be designed according to the required spatial resolution, usually between 10 and 30 centimeters. The audio terminal device is activated and set to the real-time recording mode. An appropriate sampling rate (such as 44.1 kHz or 48 kHz) should be selected to ensure sound quality. A high-quality audio coding format (such as WAV or FLAC) is used for storage to avoid losses caused by audio compression. The real-time collected sound signals need to be stored on a local or remote server. A real-time data stream processing tool (such as Apache Kafka) is used to transfer the real-time data to a database or a data warehouse for subsequent analysis. Machine learning algorithms are used for noise source identification. Deep learning models such as convolutional neural networks (CNNs) or long short-term memory networks (LSTMs) are selected, which are suitable for processing time-series audio data. Features are extracted from the collected audio signals, including Mel-frequency cepstral coefficients (MFCCs), spectrograms, pitches, etc. These features can effectively characterize the time-frequency characteristics of sounds and help improve the recognition accuracy. A labeled dataset (containing audio samples of different environmental noise sources) is used to train the model. After training, the real-time collected sound signals are input into the model for noise source identification. When the model identifies a specific noise source, the characteristics and occurrence time of the noise source are recorded and marked in the data record. The generated marks should include the noise source type (such as traffic noise, mechanical noise, natural sound, etc.) and the corresponding timestamp. The time difference of arrival (TDOA) method is selected. This method requires at least three microphones at specific positions to calculate the position of the noise source through the time difference of sound propagation. By recording the time when each microphone receives the sound, the time difference between different microphones is calculated. It is achieved through the following formula: Δt = d / c, where d is the distance from the noise source to the microphone and c is the speed of sound (calculated at 343 m / s). Using the principle of three-dimensional geometry, the three-dimensional spatial coordinates (x, y, z) of the noise source are solved through the time difference data of multiple microphones. This process is achieved through nonlinear least squares, constructing equations based on the time differences and solving them. The spatial position coordinates, noise type, intensity, etc. of each noise source are organized into a dataset. Ensure the integrity of the data for subsequent analysis. Visualization tools (such as Matplotlib, Plotly in Python, or dedicated GIS software) are used to draw the three-dimensional distribution map of the noise sources. Appropriate coordinate systems and units are selected to ensure the clarity of the graph. Draw the three-dimensional distribution map of the noise sources and analyze the spatial characteristics of the noise sources.Calculate the density distribution of noise sources, and analyze the aggregation degree and distribution uniformity of noise sources. Calculate the statistical characteristics of noise sources, such as the mean value, standard deviation, and distribution range of noise intensity, and reveal the relationship between the spatial distribution characteristics and intensity distribution of environmental noise. The normal distribution model can be used to fit the noise intensity data and analyze its distribution characteristics. Select the Kriging interpolation method to construct the environmental noise distribution field. This method can effectively process spatial data and provide a smooth noise distribution result. Input the sorted spatial positions and intensity data of noise sources into the Kriging interpolation model, calculate the noise intensity at any point, and generate a global noise distribution map. This process usually requires setting a reasonable model parameter (such as a variogram). Use independent measured data to verify the effectiveness of the constructed noise distribution model. Compare the predicted noise distribution of the model with the actual measurement data, and calculate the accuracy index of the model (such as the root mean square error RMSE).

[0023] In this embodiment, the specific steps of step S13 are as follows: Calculate the covariance matrix of the real-time environmental sound signal; Perform eigenvalue decomposition on the covariance matrix to obtain the signal subspace vector and the noise subspace vector; Perform frequency-azimuth plane traversal calculation based on the noise subspace vector to obtain the noise vector orthogonality values for different frequency and azimuth combinations; Use the noise vector orthogonality value as the MUSIC spectrum value to plot the frequency-azimuth plane to construct the MUSIC spectrum; Calculate the spectral peak value of the MUSIC spectrum and mark the MUSIC spectrum peak; Calculate the azimuth angle and elevation angle of the MUSIC spectrum peak; Determine the azimuth and elevation angles of the environmental noise source node based on the azimuth angle and elevation angle; Perform spatial azimuth positioning based on the azimuth and elevation angles of the environmental noise source node to obtain the spatial azimuth information of the noise source; Calculate the acoustic wave transmission time of the environmental noise source node to obtain the noise source distance; Perform three-dimensional precise position positioning based on the spatial azimuth information and noise source distance of the noise source to obtain the spatial position coordinates of the noise source; Repeat the above calculation steps to obtain the spatial position coordinates of each noise source.

[0024] In this embodiment, a multi-channel microphone array is used to collect environmental sound signals in real time. Ensure that the collected data has sufficient time length and sampling frequency (such as 16 kHz) for subsequent analysis. Preprocess the sound signals of each channel, including denoising and removing the DC component. Then, organize the signals of all channels into a matrix X, where each row corresponds to a time sample and each column corresponds to a microphone channel. The covariance matrix R is calculated by the following formula: R= , where N is the number of samples, and the calculated covariance matrix R will contain the statistical characteristics of the signal. Use a linear algebra library (such as numpy.linalg.eig in NumPy) to perform eigenvalue decomposition on the covariance matrix R to obtain eigenvalues λ and corresponding eigenvectors V. The eigenvalues represent the energy distribution of the signal and noise. Arrange the eigenvalues in descending order to distinguish the signal subspace and the noise subspace. Set a threshold (such as 50% of the eigenvalues) to distinguish the main signal and noise components. The eigenvectors corresponding to the first k largest eigenvalues form the signal subspace while the remaining eigenvectors form the noise subspace , define the frequency range (such as 0 Hz to 5 kHz) and the azimuth range (such as 0 to 360 degrees). For each frequency and azimuth combination, calculate the orthogonality of the noise vectors. The orthogonality of the noise vectors is calculated by the following formula: , where is the steering vector of the sound source, It is the eigenvector of the noise subspace. The calculated orthogonality value of the noise vectors is used as the MUSIC spectrum value and plotted on the frequency-azimuth plane. Usually, a pseudocolor map is used to display different spectrum values for more intuitive observation. The generated MUSIC spectrum is visualized through image processing tools (such as Matplotlib). Peak detection is performed on the MUSIC spectrum, and methods such as local maximum detection or signal processing libraries (such as the find_peaks function in SciPy) are used for calculation to obtain the peak positions of the spectrum values. The peak positions of the MUSIC spectrum are marked, and the corresponding frequencies and azimuth angles are recorded. According to the peak positions, the corresponding azimuth angles and elevation angles are calculated. The azimuth angle is usually directly obtained from the azimuth angle of the maximum point of the spectrum value, while the elevation angle is calculated through the geometric relationship of the sound source model. Based on the calculated azimuth angle and elevation angle, the azimuth and pitch angles of the environmental noise source are determined. These parameters provide necessary information for subsequent spatial positioning. The azimuth and pitch angles of each noise source are recorded for subsequent processing and analysis. Based on the obtained azimuth angle and elevation angle, three-dimensional spatial positioning of the noise source is performed. This usually requires combining the geometric layout of the microphone array and the acoustic wave propagation model. Using the triangulation method, the measurement data of multiple microphone arrays are used to improve the positioning accuracy. Using the known speed of sound (about 343 m / s in air) and the arrival time of the noise signal, the propagation time of the acoustic wave from the noise source to the microphone is calculated. According to the propagation time, the distance of the noise source is calculated, d = v⋅t. Combining the spatial azimuth information and distance of the noise source, the coordinates of the noise source in three-dimensional space are calculated.

[0025] x = d⋅sin(θ)⋅cos(ϕ) y = d⋅sin(θ)⋅sin(ϕ) z = d⋅cos(θ) where θ is the elevation angle and ϕ is the azimuth angle.

[0026] Repeat the above calculation steps to analyze multiple noise sources. Each time a new peak is detected, the corresponding azimuth angle, elevation angle, and distance are calculated to obtain the position coordinates of all noise sources. The coordinates of each noise source are stored in a database for subsequent analysis and visualization.

[0027] In this embodiment, refer to Figure 3 , which is a schematic diagram of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Step S21: Identify the spatial propagation paths according to each environmental noise source node, and extract multiple spatial propagation paths of each noise source; Step S22: Calculate the noise intensity for each environmental noise source node to generate the signal intensity of each noise source; Step S23: Perform attenuation analysis on each of the multiple spatial propagation paths of each noise source according to the signal intensity of each noise source, so as to generate the attenuation intensity distribution data of each path of the noise source; Step S24: Perform current scene noise change analysis on the real-time environmental sound signal to obtain the current scene noise change characteristics; Step S25: Dynamically render the environmental noise distribution field based on the attenuation intensity distribution data of each path of the noise source and the current scene noise change characteristics, and construct a dynamic noise scene model.

[0028] In this embodiment, the three-dimensional coordinates of each noise source node (obtained through the aforementioned steps) are determined. Using these coordinates, a spatial propagation model from the noise source to the receiving point (such as a microphone array) is constructed. A suitable propagation model is selected, such as a free-field propagation model or a complex environment model with reflections and scattering. The ray-tracing method is used to simulate the propagation path of sound waves. Considering obstacles in the environment (such as walls, furniture, etc.), a path recognition algorithm (such as the A* algorithm or Dijkstra algorithm) is used to identify multiple propagation paths from the noise source to the receiving point. This process requires first constructing a geometric model of the environment, including the position information of the noise source, obstacles, and receiving point. The characteristic information of each path is recorded, including path length, passing obstacles, reflection surfaces, and propagation loss, etc. The multiple propagation paths and their characteristic information of each noise source are stored in a database for subsequent analysis and use. The microphone array is used to collect environmental sound signals in real time, ensuring that the sampling frequency is high enough (such as 16 kHz) to obtain accurate noise signals. For each noise source node, an energy measurement method is used to calculate the noise intensity. The collected signals are preprocessed, including noise removal and DC component removal. The fast Fourier transform (FFT) is used to extract the spectral characteristics of the signals, and the intensity of each frequency component is calculated. According to the signal intensity of the noise source and the path characteristics, a path attenuation model is established. Path attenuation is usually related to distance and environmental factors (such as sound absorption and reflection). The following attenuation model is adopted: A(d)=L0−20log10(d)−αd, where A(d) is the attenuated signal intensity, L0 is the reference intensity, d is the path length, and α is the environmental attenuation coefficient. Each propagation path is analyzed for attenuation one by one, combining the noise intensity calculation with the path length and environmental attenuation to generate the attenuation intensity distribution data of each path. A simulation or simulation tool (such as MATLAB or Python) is used to implement path attenuation analysis, recording the signal intensity changes of each path at different distances. Time-frequency analysis methods such as the short-time Fourier transform (STFT) or wavelet transform are used to analyze the environmental sound signals collected in real time, extracting the characteristics of the noise signals, such as spectral characteristics, time-domain characteristics, and statistical characteristics (mean, variance, etc.). By analyzing these characteristics, the noise change situation of the current scene is identified. The noise characteristics of the current scene are compared with the previous baseline noise characteristics, analyzing the change amplitude and trend, and identifying the specific characteristics of the noise change (such as burst noise, periodic noise, etc.). According to the path attenuation intensity distribution data of each noise source and the noise change characteristics of the current scene, a dynamic noise scene model is constructed and visualized using computer graphics technology (such as OpenGL or Unity). Combining the path attenuation data and the scene noise change characteristics, an interpolation algorithm (such as Kriging interpolation or inverse distance weighting) is used to generate a noise distribution field, dynamically displaying the intensity distribution of the noise source at different time and space positions, and visualizing the noise distribution in a three-dimensional scene.Use color coding to represent the noise intensity (e.g., light color indicates low noise and dark color indicates high noise), ensure that the dynamic model can be updated in real time to reflect the noise changes in the current scene, output the dynamic noise scene model, and perform visual display, record the key parameters of the model for subsequent analysis and optimization of noise control measures.

[0029] In this embodiment, refer to Figure 4 , which is a schematic diagram of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Step S31: Perform multi-temporal noise propagation evolution on the dynamic noise scene model to generate multi-temporal noise propagation data; Step S32: Perform multi-scale time-frequency decomposition on the multi-temporal noise propagation data to obtain noise frequency domain data of different frequencies; Step S33: Calculate the amplitude of each noise signal one by one according to the noise frequency domain data of different frequencies to generate the signal amplitude value of each noise; Step S34: Perform time-frequency phase analysis on the noise frequency domain data of different frequencies to obtain the noise time-frequency phase characteristics; Step S35: Calculate the hedging suppression parameters based on the signal amplitude value of each noise and the noise time-frequency phase characteristics, so as to obtain the dynamic hedging suppression parameters of each noise; Step S36: Perform adaptive noise suppression processing based on the dynamic hedging suppression parameters of each noise to construct an adaptive noise suppression strategy.

[0030] In this embodiment, based on the existing dynamic noise scenario model, the length of the time series and the sampling frequency (the sampling frequency for 1 second is 48 kHz) are set. The time step is determined. For example, noise data is recorded every 0.1 seconds. A multi-temporal noise propagation model is constructed using computer simulation software (such as MATLAB or Python). The propagation of environmental noise in different time periods is simulated, considering the influence of time factors on the noise intensity, frequency components, and propagation path. By setting different environmental conditions (such as wind speed, temperature, obstacle position, etc.), the propagation evolution of the noise source under these conditions is simulated. According to the environmental changes, an acoustic propagation model (such as the geometric acoustics model or the wave acoustics model) is used to calculate the changes in noise in space. The noise intensity, frequency distribution, and propagation path at each time point are recorded to generate multi-temporal noise propagation data. These data will be used for subsequent analysis. The generated multi-temporal noise propagation data is stored in a database, including timestamps, noise intensity, frequency components, and interference information, ensuring that the data structure is clear and easy to access. An appropriate time-frequency analysis method is selected, such as the Wavelet Transform or the Short-Time Fourier Transform (STFT), to extract the frequency characteristics in different time periods. Specific parameters are set, such as the type of wavelet function (such as the Morlet wavelet) and the scale range, to ensure coverage of the frequency range of interest (such as 20 Hz to 20 kHz). The selected time-frequency decomposition algorithm is applied to the noise data at each time point, and the calculated time-frequency representation reveals the changes in different frequency components over time. The pywt library in Python is used for wavelet transform to perform multi-scale decomposition on the noise signal and extract the noise frequency domain data of different frequencies. For each frequency component, its corresponding signal amplitude is calculated. This is usually done by calculating the root mean square (RMS) value of each frequency component or directly extracting the amplitude value from the spectrum. The FFT is used to calculate the spectrum within each time period, and the amplitude of each frequency component is extracted. In MATLAB or Python, the numpy.fft module is used for FFT analysis, and the amplitude values of each noise signal are recorded in the database and associated with the timestamps and frequency information to ensure the integrity and traceability of the data. For each frequency component, its phase information is extracted. Phase information is crucial for understanding the propagation characteristics and interference effects of sound waves. Using the results of the FFT, the phase values of each frequency component are extracted: According to the amplitude and phase characteristics of the noise signal, the hedge suppression parameter is calculated. The hedge suppression technique aims to reduce noise interference and improve signal quality. An optimization algorithm (such as the least squares method or the genetic algorithm) is used to calculate the optimal hedge suppression parameter. A target function is set to minimize the difference between the noise amplitude and the signal amplitude. The signal amplitude values and phase characteristics of each frequency are input into the optimization model to calculate the hedge suppression parameters. These parameters include gain, phase compensation, etc., and are used for subsequent noise suppression processing. An adaptive noise suppression algorithm is designed based on the dynamic hedge suppression parameters of each noise source.Common methods include adaptive filters (such as LMS or RLS algorithms). In real-time signal processing, an adaptive filter is used to process the input signal, and the hedge suppression parameters obtained from the aforementioned calculations are used to adjust the gain and phase of the filter. Implement the adaptive filter in MATLAB or Python to update the filter coefficients in real time to reduce noise interference and improve signal quality. Evaluate the quality of the processed signal by comparing the signal strength and clarity before and after processing, and use metrics such as signal-to-noise ratio (SNR) for evaluation. Record the results of the adaptive noise suppression process and feedback them to the system for further optimization of the algorithm and parameter settings.

[0031] In this embodiment, step S4 includes the following steps: Step S41: Obtain the audio input signal of the user; perform in-depth audio semantic parsing on the audio input signal of the user to extract the user audio semantic features; Step S42: Perform key semantic recognition on the user audio semantic features to extract the user key semantic audio data; Step S43: Perform local audio gain processing on the user key semantic audio data to obtain a local gain audio signal; Step S44: Perform multi-band decomposition on the local gain audio signal to generate gain audio signals in multiple bands; Step S45: Calculate the audio frequency for each band of the gain audio signals in multiple bands to generate the audio frequency parameters for each band.

[0032] In this embodiment, a high-quality microphone is used to collect the user's audio input signal, ensuring that the sampling rate is at least 16 kHz to guarantee audio quality. Real-time recording is performed using audio recording software (such as Audacity) or a programming environment (such as the PyAudio library in Python). During the recording process, ensure that the environment is quiet to reduce background noise interference, and at the same time use noise suppression technology to further improve the signal quality. The collected audio signal is input into a deep learning model, such as an audio processing model based on a convolutional neural network (CNN) or a recurrent neural network (RNN). These models perform feature extraction and semantic analysis on the audio signal. Use a pre-trained model (such as Google's Speech-to-Text API or a custom deep learning model) to parse the audio signal and extract the semantic features of the audio. These features include information such as pitch, volume, emotion, pronunciation, and speech rate. Build or use an existing key semantic recognition model, usually based on natural language processing (NLP) technology. Adopt a deep learning model (such as BERT or GPT) to identify the key semantics in the audio. Define the recognition criteria for key semantics, such as setting a threshold to identify words and phrases with a higher frequency or stronger emotion. Input the extracted audio semantic features into the key semantic recognition model, and the model will analyze the audio content and return the key semantic data. Use text mining techniques to identify the user's main intention or emotional state. Record the identified key semantics and their timestamps for subsequent processing. Store the key semantic audio data (such as text and its corresponding audio segments) in a database and ensure its association with the original audio signal for subsequent analysis and traceability. To highlight the user's key semantic audio data, use audio gain processing technology, such as dynamic range compression (DRC). This technology can increase the overall volume of the audio signal without distortion. Use an audio processing library (such as Librosa or Soundfile) to perform gain processing on the key semantic audio data. First, identify the time periods of key semantics in the audio signal, and then adjust the gain of these segments. Set the gain value (such as 3–6 dB) and process the key semantic segments to ensure that the enhanced audio signal is clearly audible. Save the processed local gain audio signal as a new audio file (such as WAV or MP3 format), and record the processing parameters (gain value and processing time) for subsequent review and adjustment. Use a filter bank (such as a Butterworth filter or wavelet transform) to perform multi-band decomposition on the local gain audio signal. The purpose of this step is to decompose the audio signal into sub-signals in multiple frequency ranges for more detailed processing. Implement multi-band decomposition using the scipy.signal library in Python. Set the parameters of the filter, such as the cut-off frequency and bandwidth, usually selecting frequency bands such as low frequency (20 Hz–250 Hz), middle frequency (250 Hz–2 kHz), and high frequency (2 kHz–20 kHz).Apply these filters to the audio signal after gain processing to extract the gain audio signals in each frequency band. Save the gain audio signal of each band as a separate audio file, and record the frequency range and gain characteristics of each band for subsequent analysis and processing. For each extracted band, use the Fast Fourier Transform (FFT) to calculate its frequency characteristics. FFT effectively converts the time-domain signal into a frequency-domain signal, thereby obtaining the frequency parameters of each band. Use the numpy.fft library in Python to perform FFT analysis on the gain signal of each band. Calculate the spectrum of each band, and extract the main frequency components and corresponding amplitude values. Record the main frequency and its amplitude of each band, usually selecting the first few peak frequencies for recording.

[0033] In this embodiment, step S5 includes the following steps: Step S51: Detect the noise resonance frequency of the dynamic noise scenario model according to the audio frequency parameters of each band, and extract the audio resonance frequency band; Step S52: Perform available frequency adjustment calculation on the audio resonance frequency band to generate the available frequency adjustment range of the resonance band; Step S53: Perform optimal frequency analysis based on the available frequency adjustment range of the resonance band to generate the optimal adjustment frequency; Step S54: Perform dynamic frequency fine-tuning on the audio resonance frequency band according to the optimal adjustment frequency to generate the dynamically frequency fine-tuned audio band; Step S55: Perform frequency adjustment distortion analysis on the dynamically frequency fine-tuned audio band to generate the audio distortion characteristics; Step S56: Perform non-linear distortion compensation calculation on the audio distortion characteristics to generate the non-linear distortion compensation parameters; Step S57: Use the non-linear distortion compensation parameters to perform immediate distortion compensation processing on the dynamically frequency fine-tuned audio band, thereby obtaining the distortion compensation audio band; Step S58: Based on the distortion compensation audio band, perform global reconstruction on the local gain audio signal to construct the globally reconstructed and optimized audio signal.

[0034] In this embodiment, in order to identify the noise resonance frequency, it is first necessary to obtain the frequency data in the dynamic noise scenario model. The noise scenario is analyzed using the Short-Time Fourier Transform (STFT) to extract the frequency characteristics within each time period. A frequency range, such as 20 Hz to 20 kHz, is set as the target range for detection. The FFT is applied to the dynamic noise scenario model to obtain the spectral information. In Python, the numpy.fft.fft function is used to calculate the spectrum. The resonance frequency is identified by setting a threshold (the signal amplitude is greater than a certain value). All frequencies that meet the conditions are recorded and classified into the audio resonance frequency band. For the detected audio resonance frequency band, the calculation of available frequency adjustment is performed. This process needs to consider the environmental adaptability and sound quality requirements of each band. Determine the current frequency range of each resonance frequency band and set an available adjustment range for each band. According to actual requirements, such as a frequency offset of ±5% or ±10%, set the upper and lower limits of the frequency adjustment. Calculate the available frequency adjustment range and generate the corresponding frequency adjustment parameters. Record the available frequency adjustment range of each resonance band in the database, ensuring that it is associated with the corresponding frequency band information for subsequent analysis. For the available frequency adjustment range of each resonance band, the optimal frequency analysis is performed. The optimal frequency usually refers to the frequency that can maximize the audio signal quality or response characteristics. Within the available frequency range, an optimization algorithm (such as a genetic algorithm or particle swarm optimization) is used to determine the optimal frequency. Set the objective function, such as maximizing the signal intensity or minimizing the distortion. In Python, the scipy.optimize library is used for the optimization calculation to find the best frequency. Store the optimal adjustment frequency of each resonance band in the database and record the parameters and results used in the optimization process for subsequent analysis and adjustment. For the optimal adjustment frequency of each resonance frequency band, the dynamic frequency fine-tuning is performed. This process is achieved through a digital filter. A digital filter (such as a Butterworth or Chebyshev filter) is used to process the audio signal to adjust the signal to the optimal frequency. Set the parameters of the filter (such as the cut-off frequency and gain) to adapt to the optimal adjustment frequency. In Python, the scipy.signal library is used to implement the design and application of the digital filter. The frequency adjustment distortion analysis aims to evaluate the distortion degree of the audio band after fine-tuning. Metrics such as THD (Total Harmonic Distortion) and IMD (Intermodulation Distortion) are used for evaluation. Calculate the spectrum of the adjusted audio band and compare it with the original band. Use the FFT to obtain the spectral data and calculate the distortion metrics. Set a threshold (such as 5%) to determine whether the distortion is acceptable. Use the numpy library in Python for the distortion calculation and analysis. Record the distortion analysis results in the database, ensuring that detailed information about the distortion characteristics is included for subsequent distortion compensation calculations. For the detected distortion characteristics, the non-linear distortion compensation calculation is performed. Non-linear distortion compensation usually relies on dynamic range compression and phase compensation techniques.Design corresponding compensation algorithms according to the distortion characteristics. Set the parameters of the dynamic range compressor to reduce distortion, or apply phase compensation techniques to adjust the signal phase. Use the signal processing library in Python to design and implement the non-linear distortion compensation processing flow. Store the calculated non-linear distortion compensation parameters in the database, and record in detail the algorithms and processing results used for subsequent applications. Implement real-time distortion compensation processing according to the calculated non-linear distortion compensation parameters. This processing should be carried out during the real-time transmission of the signal to ensure the quality of the audio signal. Apply the non-linear distortion compensation algorithm to the audio band after dynamic frequency fine-tuning, and adjust the audio signal in real time to compensate for the distortion. Use an adaptive filter to dynamically adjust the signal. Implement this algorithm in the audio processing software to ensure that the processing speed can meet the real-time requirements, usually requiring a delay of less than 20 milliseconds. Save the distorted-compensated audio band as a new audio file, and record the parameters and processing effects used during the compensation for subsequent analysis. After the distortion compensation processing is completed, perform global reconstruction. The global reconstruction aims to combine the compensated audio signal with the local gain audio signal to improve the overall quality of the audio. Use weighted average or hybrid techniques to merge the distorted-compensated audio band with the local gain audio signal. Set appropriate weights to ensure that the contribution of each signal is reasonably reflected. Use the audio processing library in Python to reconstruct the signal and generate the final optimized audio signal.

[0035] In this embodiment, step S6 includes the following steps: Step S61: Perform multi-scenario transfer learning on the globally reconstructed optimized audio signal to generate multi-scenario learning samples; Step S62: Optimize the audio noise reduction parameters of the multi-scenario learning samples to generate an intelligent audio noise reduction optimization strategy; Step S63: Dynamically iteratively optimize the intelligent audio noise reduction optimization strategy and the adaptive noise suppression strategy to construct an intelligent multi-scenario noise reduction engine.

[0036] In this embodiment, the goal of multi-scenario transfer learning is defined. The core idea of transfer learning is to utilize the knowledge obtained from training on one scenario to improve the learning effect on a new scenario. The audio signal is selected as the object of transfer learning and implemented through a deep learning model (such as CNN or LSTM). The types of scenarios to be processed are determined (such as indoor, outdoor, noisy environment, quiet environment, etc.), and corresponding audio samples are collected for each scenario. The audio signal is optimized based on global reconstruction to generate multi-scenario learning samples. The librosa library is used in Python to read and process audio data to ensure the quality and diversity of the samples. Data augmentation is performed to simulate different environmental conditions, such as adding background noise, changing the audio speed and pitch, etc., to expand the sample set. The audiomentations library is used to achieve this. The prepared multi-scenario samples are used to train the deep learning model. A suitable loss function (such as mean squared error) and optimizer (such as Adam) are selected, and the learning rate (such as 0.001) is set for model training. Cross-validation is performed on each scenario to evaluate the model performance and ensure that the model can be effectively transferred to different scenarios. The loss and accuracy of each training are recorded for subsequent analysis and adjustment. Based on the multi-scenario learning samples, the key parameters affecting the audio noise reduction effect are identified. These parameters include noise type, signal strength, frequency response, etc. Statistical learning methods (such as regression analysis) are used to evaluate the impact of these parameters on the noise reduction effect to determine the objectives to be optimized. An optimization algorithm is designed, such as particle swarm optimization (PSO) or genetic algorithm, to optimize the noise reduction parameters. These algorithms search in the parameter space to find the best combination of noise reduction parameters. The genetic algorithm is implemented in Python using the deap library, setting the population size (such as 50) and the number of iterations (such as 100) to ensure convergence to the optimal solution. The optimized noise reduction parameters are verified by testing with audio signals from different scenarios, and the change in signal-to-noise ratio (SNR) before and after noise reduction is recorded. The optimized noise reduction parameters and strategies are stored in the database to provide a basis for the subsequent intelligent audio noise reduction engine. A framework for dynamic iterative optimization is designed, combining the optimization strategy with an adaptive noise suppression strategy. This framework can monitor the changes in the audio signal in real time and dynamically adjust the noise reduction parameters according to the changes in the environmental noise. An online learning algorithm (such as stochastic gradient descent) is used to dynamically adjust the noise reduction strategy. A threshold is set, for example, when the SNR is below a certain value, the noise reduction parameters are automatically adjusted to adapt to the current environment. A feedback mechanism is implemented in Python, inputting the characteristics of the real-time audio signal (such as frequency distribution and noise intensity) into the optimization model to update the noise reduction strategy. Performance tests are carried out under different scenarios to evaluate the effect of the intelligent multi-scenario noise reduction engine. The real-time noise reduction effect and user feedback are recorded, and A / B testing is used to compare the performance of different noise reduction strategies. The algorithm is continuously optimized according to the test results to improve the adaptability and response speed of the noise reduction engine.

[0037] In this embodiment, a multi-scenario adaptive noise reduction system for an audio terminal device is provided, which is used to execute the multi-scenario adaptive noise reduction method for the audio terminal device as described above, and includes: A noise perception module, which is used to identify real-time environmental sound signals; identify environmental noise sources for the real-time environmental sound signals, and perform noise distribution perception modeling to construct an environmental noise distribution field; A noise distribution rendering module, which is used to analyze the current scene noise change of the environmental noise distribution field, and perform dynamic noise distribution rendering to construct a dynamic noise scene model; A noise suppression module, which is used to calculate hedge suppression parameters based on the dynamic noise scene model, and perform adaptive noise suppression processing to construct an adaptive noise suppression strategy; A local audio gain module, which is used to obtain the user's audio input signal; perform local audio gain processing on the user's audio input signal, and perform multi-band decomposition to generate audio frequency parameters for each band; A resonance frequency fine-tuning module, which is used to fine-tune the noise resonance frequency of the dynamic noise scene model according to the audio frequency parameters of each band, and perform instantaneous distortion compensation processing to construct a globally reconstructed and optimized audio signal; An intelligent noise reduction optimization module, which is used to perform multi-scenario transfer learning on the globally reconstructed and optimized audio signal, and perform dynamic iterative optimization according to the adaptive noise suppression strategy to construct an intelligent multi-scenario noise reduction engine.

[0038] Through the use of multimodal input from a microphone array and sensors, the present invention can comprehensively collect sound signals in the current environment. This lays a foundation for subsequent noise source localization, classification, and processing. By analyzing the audio characteristics of different noise sources in the environment (such as frequency, intensity, time-domain variation, etc.), the system can identify the types of noise sources in the environment (such as wind noise, traffic noise, conversation sounds, etc.). Through the application of an acoustic model, the device can establish a noise distribution field to help understand the distribution of noise sources in space, facilitating subsequent noise suppression and optimization processing. Through dynamic noise analysis, the device can continuously monitor the changes in noise in the environment and automatically adapt to the noise changes in different scenarios. During commuting, the noise sources are constantly changing, and noise sources such as vehicles passing by and road construction are constantly switching. The system can accurately identify these changes and adjust the noise reduction strategy. By performing time-varying modeling on the noise source, the device can construct a dynamic noise scenario model to adapt to the instantaneous changing environmental noise. This not only enhances the accuracy of the noise reduction algorithm but also provides accurate noise source location and noise type data for subsequent adaptive noise reduction algorithms. By calculating the dynamic noise scenario model, the device can calculate effective noise cancellation suppression parameters based on the intensity, position, and propagation path of the noise source. This cancellation suppression method reduces the interference of noise on the audio signal by generating a sound waveform that opposes the noise source. The device can automatically adjust the noise suppression strategy according to the changes and characteristics of the noise in the current environment. Compared with traditional fixed-parameter noise reduction methods, adaptive noise suppression can make precise adjustments according to the noise intensity, spectrum, and scene requirements, ensuring the best noise reduction effect in different environments. As the environmental noise changes, the system can dynamically adapt to these changes, not only reducing common noises (such as traffic noise, wind noise, etc.) but also coping with instantaneous noises (such as sudden noises or human voice interference). Through local gain processing of the user's audio signal, the clarity and quality of the user's speech can be enhanced, avoiding speech distortion caused by excessive noise suppression. At the same time, the multi-band decomposition technology enables the device to precisely process audio signals in different frequency bands, thereby effectively distinguishing speech from noise. After generating frequency parameters for each audio band, the device can more precisely identify and suppress noise according to the characteristics of different frequencies. Different processing strategies are adopted for low-frequency noise and high-frequency noise, and this frequency band decomposition enables the device to flexibly cope with diverse noise environments. According to the frequency parameters of each audio band, the system can detect and adjust the resonance frequency close to the frequency of the noise source. This step can optimize the spectrum of the audio signal, effectively suppressing the noise in the resonance frequency band while preserving the naturalness of the speech. By performing distortion compensation on the audio signal, the system can correct the sound quality loss caused by the noise reduction process. This immediate compensation ensures that the device maintains the clarity and naturalness of the audio during the noise suppression process.By optimizing the synthesis of noise for each band, the device can output a globally reconstructed audio signal, ensuring that users receive high-quality and clear sound, whether in a quiet environment or a noisy scene. The device can learn effective noise reduction strategies from different noise scenarios through transfer learning technology and apply them to new scenarios. Transfer learning enables the system to be adaptable from the initial stage, reducing training time and improving noise reduction efficiency. According to environmental changes and user needs, the device will continuously self-learn and optimize to ensure optimal noise reduction effects are continuously provided in multiple scenarios. When the user enters a noisy public transportation vehicle or a meeting, the device can automatically identify the scenario and adjust the noise reduction strategy. Through continuous iteration and optimization, the system can adjust the noise reduction parameters in real time in different scenarios, avoiding the static setting of traditional noise reduction methods, enabling precise noise suppression in each environment and enhancing the user's audio experience.

[0039] Therefore, in any regard, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be encompassed within the present invention.

[0040] As described above, these are only specific embodiments of the present invention to enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein but will conform to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A multi-scenario adaptive noise reduction method for an audio terminal device, characterized in that: The following steps are involved: Step S1: identifying a real-time environmental sound signal; identifying an environmental noise source for the real-time environmental sound signal, and performing noise distribution perception modeling to construct an environmental noise distribution field; Step S2: Analyze the current scene noise change of the environmental noise distribution field, perform dynamic noise distribution rendering, and construct a dynamic noise scene model; Step S3: Calculate the hedging suppression parameters based on the dynamic noise scene model, perform adaptive noise suppression processing, and construct an adaptive noise suppression strategy; Step S4: obtaining the user's audio input signal; performing local audio gain processing on the user's audio input signal, and performing multi-band decomposition to generate audio frequency parameters for each band; Step S5: fine-tuning the noise resonance frequency of the dynamic noise scene model according to the audio frequency parameters of each band, and performing real-time distortion compensation processing to construct a global reconstruction optimized audio signal; Step S6: Perform multi-scenario transfer learning on the globally reconstructed optimized audio signal, and perform dynamic iterative optimization according to the adaptive noise suppression strategy to build an intelligent multi-scenario noise reduction engine.

2. The multi-scenario adaptive noise reduction method of the audio terminal device according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: recognizing real-time environmental sound signals based on a multi-sensor array of an audio terminal device; Step S12: Identify the environmental noise source of the real-time environmental sound signal and mark each environmental noise source node; Step S13: performing three-dimensional spatial positioning on each environmental noise source node to obtain the spatial position coordinates of each noise source; Step S14: performing environmental noise distribution analysis on the spatial position coordinates of each noise source to obtain spatial distribution characteristics of environmental noise; Step S15: Perform noise distribution perception modeling according to the spatial distribution characteristics of the ambient noise to construct the ambient noise distribution field.

3. The multi-scenario adaptive noise reduction method of the audio terminal device according to claim 2, characterized in that: The specific steps of step S13 are: Calculating the covariance matrix of the real-time ambient sound signal; Perform eigenvalue decomposition on the covariance matrix to obtain a signal subspace vector and a noise subspace vector; Perform frequency-azimuth plane traversal calculation according to the noise subspace vector to obtain orthogonality values ​​of noise vectors of different frequency and azimuth combinations; According to the orthogonality value of the noise vector as the MUSIC spectrum value, the frequency-azimuth plane is plotted to construct the MUSIC spectrum; Calculate the peak value of the MUSIC spectrum and mark the peak value of the MUSIC spectrum; Calculating the azimuth and elevation of the MUSIC spectrum peak; Determine the azimuth and elevation angles of the environmental noise source node based on the azimuth and elevation angles; Performing spatial orientation positioning based on the azimuth and elevation angle of the environmental noise source node, thereby obtaining spatial orientation information of the noise source; Calculating the sound wave transmission time of the environmental noise source node to obtain the noise source distance; Based on the spatial orientation information and distance of the noise source, three-dimensional precise positioning is performed to obtain the spatial position coordinates of the noise source; Repeat the above calculation steps to obtain the spatial position coordinates of each noise source.

4. The multi-scenario adaptive noise reduction method of the audio terminal device according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: performing spatial propagation path identification according to each environmental noise source node, and extracting multiple spatial propagation paths of each noise source; Step S22: Calculate the noise intensity of each environmental noise source node to generate the signal strength of each noise source; Step S23: performing path attenuation analysis on multiple spatial propagation paths of each noise source one by one according to the signal strength of each noise source, so as to generate attenuation intensity distribution data of each path of the noise source; Step S24: Analyze the current scene noise change of the real-time environmental sound signal to obtain the current scene noise change characteristics; Step S25: Based on the attenuation intensity distribution data of each path of the noise source and the noise change characteristics of the current scene, the environmental noise distribution field is dynamically rendered to construct a dynamic noise scene model.

5. The multi-scenario adaptive noise reduction method of the audio terminal device according to claim 1, characterized in that: The specific steps of step S3 are: Step S31: performing multi-time series noise propagation evolution on the dynamic noise scene model to generate multi-time series noise propagation data; Step S32: performing multi-scale time-frequency decomposition on the multi-time series noise propagation data to obtain noise frequency domain data of different frequencies; Step S33: Calculating the noise signal amplitude one by one according to the noise frequency domain data of different frequencies to generate the signal amplitude value of each noise; Step S34: performing time-frequency phase analysis according to the noise frequency domain data of different frequencies to obtain the noise time-frequency phase characteristics; Step S35: Calculate the hedging suppression parameter based on the signal amplitude value and the noise time-frequency phase characteristics of each noise, so as to obtain the dynamic hedging suppression parameter of each noise; Step S36: Perform adaptive noise suppression processing based on the dynamic hedging suppression parameters of each noise to construct an adaptive noise suppression strategy.

6. The multi-scenario adaptive noise reduction method of the audio terminal device according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: obtaining an audio input signal of a user; performing deep audio semantic analysis on the audio input signal of the user to extract audio semantic features of the user; Step S42: performing key semantic recognition on the user audio semantic features to extract the user key semantic audio data; Step S43: performing local audio gain processing on the user key semantic audio data, thereby obtaining a local gain audio signal; Step S44: performing multi-band decomposition on the local gain audio signal to generate gain audio signals of multiple bands; Step S45: performing audio frequency calculation on the gain audio signals of the multiple bands one by one to generate audio frequency parameters of each band.

7. The multi-scenario adaptive noise reduction method of the audio terminal device according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: performing noise resonance frequency detection on the dynamic noise scene model according to the audio frequency parameters of each band, and extracting the audio resonance frequency band; Step S52: performing available frequency adjustment calculation on the audio resonance frequency band to generate an available frequency adjustment range of the resonance band; Step S53: performing optimal frequency analysis based on the available frequency adjustment range of the resonance band to generate an optimal adjustment frequency; Step S54: dynamically fine-tuning the audio resonance frequency band according to the optimal adjustment frequency to generate a dynamic frequency fine-tuning audio band; Step S55: performing frequency adjustment distortion analysis on the dynamic frequency fine-tuning audio band to generate audio distortion features; Step S56: performing nonlinear distortion compensation calculation on the audio distortion feature to generate nonlinear distortion compensation parameters; Step S57: using the nonlinear distortion compensation parameter to perform real-time distortion compensation processing on the dynamic frequency fine-tuning audio band, thereby obtaining a distortion compensated audio band; Step S58: globally reconstructing the local gain audio signal based on the distortion compensated audio band to construct a globally reconstructed optimized audio signal.

8. The multi-scenario adaptive noise reduction method of the audio terminal device according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: performing multi-scenario transfer learning on the global reconstruction optimized audio signal to generate multi-scenario learning samples; Step S62: Optimizing audio noise reduction parameters of multi-scenario learning samples to generate an intelligent audio noise reduction optimization strategy; Step S63: Dynamically iteratively optimize the intelligent audio noise reduction optimization strategy and the adaptive noise suppression strategy to build an intelligent multi-scenario noise reduction engine.

9. A multi-scenario adaptive noise reduction system for an audio terminal device, characterized in that: The multi-scenario adaptive noise reduction method for executing the audio terminal device according to claim 1 comprises: The noise perception module is used to identify real-time environmental sound signals; identify environmental noise sources for the real-time environmental sound signals, perform noise distribution perception modeling, and construct an environmental noise distribution field; The noise distribution rendering module is used to analyze the current scene noise changes in the environmental noise distribution field, perform dynamic noise distribution rendering, and construct a dynamic noise scene model; The noise suppression module is used to calculate the hedge suppression parameters based on the dynamic noise scene model, perform adaptive noise suppression processing, and build an adaptive noise suppression strategy; The local audio gain module is used to obtain the user's audio input signal; perform local audio gain processing on the user's audio input signal, and perform multi-band decomposition to generate audio frequency parameters for each band; The resonance frequency fine-tuning module is used to fine-tune the noise resonance frequency of the dynamic noise scene model according to the audio frequency parameters of each band, and perform real-time distortion compensation processing to construct a global reconstruction optimization audio signal; The intelligent noise reduction optimization module is used to perform multi-scenario transfer learning on the globally reconstructed optimized audio signal, and to perform dynamic iterative optimization based on the adaptive noise suppression strategy to build an intelligent multi-scenario noise reduction engine.

Images