Context-aware intelligent audio equalizer adaptive adjustment system
By using a scene-aware intelligent audio equalizer adaptive adjustment system, which combines a lightweight decision tree model with hardware and software filters, the system solves the problems of insufficient environmental adaptability and real-time performance of audio devices, and achieves low-power and high-efficiency audio processing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2026-03-13
- Publication Date
- 2026-06-09
AI Technical Summary
Existing audio devices are inadequate in terms of environmental adaptability and real-time performance. EQ parameter adjustment is complex and power consumption is high, making it difficult to run in real time on resource-constrained embedded terminals.
An adaptive adjustment system for an intelligent audio equalizer based on scene awareness is adopted, including audio acquisition and preprocessing, feature extraction, scene classification and dynamic equalizer configuration. It utilizes a lightweight decision tree model and a combination of hardware and software filters to achieve rapid scene recognition and dynamic EQ adjustment.
It improves the environmental adaptability and real-time performance of audio processing, reduces computational overhead and energy consumption, and is suitable for scenarios such as mobile devices and in-vehicle systems, providing an efficient and flexible audio adaptive processing solution.
Smart Images
Figure CN122179710A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a scene-aware intelligent audio equalizer adaptive adjustment system, belonging to the field of audio signal processing. Background Technology
[0002] As audiophiles increasingly demand high-quality audio equipment, their requirements for real-time audio processing performance and power consumption also rise. This necessitates the use of low-power, high-performance digital signal processor (DSP) platforms to achieve audio scene recognition and equalization adjustment, thus meeting users' sound quality needs. Existing audio devices such as headphones and speakers primarily employ the following solutions to meet these listening adjustment requirements: first, equalizer parameters (EQ parameters) are typically provided in a limited number of fixed configurations at the factory; second, user interfaces are provided for manual parameter adjustment; and third, audio characteristics are uploaded to the cloud, processed offline or online using complex models, and then the parameters are returned.
[0003] The above solution has the following shortcomings: First, adjusting EQ parameters is somewhat complex. Without a professional sound engineer, it is difficult to configure a solution with good results and to take into account the listening experience in various usage scenarios. Once the environment in which the device is located changes, there is a lack of dynamic management mechanism for the equalization band, which can lead to a decrease in sound quality or speech clarity. Secondly, cloud-based or complex deep learning model-based solutions often rely on continuous data uplink or high-computing-power inference, making it difficult to run in real time and with low power consumption on resource-constrained embedded terminals. Summary of the Invention
[0004] The purpose of this invention is to provide a scene-aware intelligent audio equalizer adaptive adjustment system to solve the problems of insufficient environmental adaptability and real-time performance, as well as high computational overhead and energy consumption in the existing technology.
[0005] The technical solution of this invention is: A scene-aware intelligent audio equalizer adaptive adjustment system includes an audio acquisition and preprocessing module, a feature extraction module, a scene classification module, and a dynamic equalizer configuration module. Audio Acquisition and Preprocessing Module: Uses a microphone to sample and acquire raw audio data in real time, and performs pre-emphasis, frame segmentation, and Hamming windowing operations on the acquired raw audio data to obtain preprocessed data; Feature extraction module: Performs frequency domain transformation on the preprocessed data, calculates acoustic feature values including Mel frequency cepstral coefficients (MFCC), spectral centroid, and zero-crossing rate, and sends the acoustic feature values to the scene classification module; Scene classification module: Inputs acoustic feature values into a lightweight decision tree model to achieve scene recognition and output scene labels; Dynamic equalizer configuration module: Adjusts the equalizer's filtering parameters, including filter gain and quality factor Q, in real time based on scene tags, and dynamically configures the number of enabled segments for hardware and software filters to achieve seamless audio effect adjustment.
[0006] Furthermore, the audio acquisition and preprocessing module includes a real-time sampling stage, a pre-emphasis stage, a framing stage, and a windowing stage, wherein: Real-time sampling stage: The raw audio data acquired by the analog-to-digital converter (ADC) is directly written to the circular buffer at a sampling rate Fs through the direct memory access DMA peripheral. Combined with the timestamp synchronization mechanism, it ensures that no frames are lost. A high-priority interrupt is triggered when the buffer is half full, thus completing jitter-free real-time sampling. Pre-emphasis stage: Perform first-order pre-emphasis filtering; In the framing stage: a dual-buffering alternation strategy based on fixed frame length and frame shift is adopted, combined with the Ping-Pong Buffer mechanism, to automatically switch the read and write areas when each frame of data arrives; Windowing stage: Apply a Hamming window to each frame of audio data to obtain windowed data and store it in a double buffer.
[0007] Furthermore, the audio acquisition and preprocessing module uses DMA circular buffer technology to achieve zero-copy transmission and is scheduled as a high-priority task in the system.
[0008] Furthermore, the feature extraction module includes a Fast Fourier Transform (FFT) calculation stage, a Mel filter group mapping stage, a Discrete Cosine Transform (DCT) calculation stage, and a parallel computing stage, wherein: Fast Fourier Transform (FFT) calculation stage: Performs fixed-point Fast Fourier Transform (FFT) operations on the preprocessed data and calculates the amplitude spectrum and power spectrum in real time; Mel filter group mapping stage: The amplitude spectrum is weighted and summed using the Mel filter weight matrix to obtain the energy value; Discrete Cosine Transform (DCT) Calculation Stage: The energy values are processed using the Discrete Cosine Transform (DCT) algorithm to obtain the transformed eigenvalues. Parallel computing stage: Three digital signal processor (DSP) parallel cores are started simultaneously to calculate acoustic characteristic values, including Mel frequency cepstral coefficients (MFCC), zero-crossing rate, and spectral centroid, and write them back to shared memory synchronously in frame order.
[0009] Furthermore, during the parallel computing phase, acoustic eigenvalues including Mel-frequency cepstral coefficients (MFCC), zero-crossing rate, and spectral centroid are calculated separately. (1) Mel frequency cepstral coefficient MFCC sub-stage: The first L terms are extracted from the transformed eigenvalues obtained from the Discrete Cosine Transform (DCT) stage as Mel frequency cepstral coefficients MFCC; (2) Zero-crossing rate sub-stage: The zero-crossing rate is obtained by rapidly accumulating the symbol change count of the preprocessed data frame by frame; (3) Spectral centroid stage: The spectral centroid is obtained by weighted summation of amplitude spectrum and frequency index.
[0010] Furthermore, in the scene classification module, acoustic feature values are input into a lightweight decision tree model to achieve scene recognition and output scene labels, specifically as follows: Step 1: Calculate the low-frequency energy ratio η based on the power spectrum obtained from the Fast Fourier Transform (FFT) calculation stage. If the low-frequency energy ratio η is less than the first set threshold, the environment is determined to be silent; otherwise, proceed to the next step 2. Step 2: Compare the zero-crossing rate ZCR with the second set threshold: If the zero-crossing rate ZCR is less than the second set threshold, it is determined to be a speech environment; otherwise, it is a noise environment, and proceed to the next step 3. Step 3: Compare the spectral centroid with the third set threshold. If the spectral centroid is higher than the third set threshold, it is determined to be a traffic noise environment; if the spectral centroid is not higher than the third set threshold, it is determined to be an indoor noise environment.
[0011] The beneficial effects of this invention are as follows: This scene-aware intelligent audio equalizer adaptive adjustment system achieves a complete process including audio acquisition, multi-dimensional feature extraction, rapid scene classification, dynamic equalizer adjustment, and low-power scheduling management. Through audio feature extraction and decision tree classification algorithms, as well as dynamic equalizer adjustment, this system can improve the environmental adaptability and real-time performance of audio processing, as well as power efficiency, and reduce computational overhead and energy consumption. It can identify different audio environments and adaptively adjust the device sound effects according to different situations. It can also fully take into account the real-time requirements and energy efficiency needs of embedded systems, providing a highly efficient and flexible audio adaptive processing solution for application scenarios such as mobile devices and in-vehicle systems, helping audio processing technology to develop towards intelligence. Attached Figure Description
[0012] Figure 1 This is a schematic diagram illustrating the scene-aware intelligent audio equalizer adaptive adjustment system according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the feature extraction module in the embodiment; Figure 3 This is a schematic diagram illustrating the scene classification module in the embodiment. Detailed Implementation
[0013] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0014] This embodiment provides a scene-aware intelligent audio equalizer adaptive adjustment system, including an audio acquisition and preprocessing module, a feature extraction module, a scene classification module, and a dynamic equalizer configuration module. Audio Acquisition and Preprocessing Module: Uses a microphone to sample and acquire raw audio data in real time. Performs pre-emphasis, frame segmentation, and Hamming windowing operations on the acquired raw audio data to obtain preprocessed data.
[0015] The audio acquisition and preprocessing module includes a real-time sampling stage, a pre-emphasis stage, a framing stage, and a windowing stage, among which: Real-time sampling stage: The raw audio data acquired by the analog-to-digital converter (ADC) is directly written to the circular buffer at a sampling rate Fs through the direct memory access DMA peripheral. Combined with the timestamp synchronization mechanism, it ensures that no frames are lost. A high-priority interrupt is triggered when the buffer is half full, thus completing jitter-free real-time sampling. Pre-emphasis stage: Perform first-order pre-emphasis filtering; In the framing stage: a dual-buffering alternation strategy based on fixed frame length and frame shift is adopted, combined with the Ping-Pong Buffer mechanism, to automatically switch the read and write areas when each frame of data arrives; Windowing stage: Apply a Hamming window to each frame of audio data to obtain windowed data and store it in a double buffer.
[0016] In the audio acquisition and preprocessing module, the pre-emphasized sequence can be framed in 20ms (320 samples) frames with 50% overlap. Hamming windows are applied to reduce spectral leakage and improve the accuracy of frequency domain analysis. The audio acquisition and preprocessing module employs DMA circular buffering technology to achieve zero-copy transmission and is scheduled as a high-priority task in the system to further ensure real-time performance.
[0017] Feature extraction module: Performs frequency domain transformation on the preprocessed data, calculates acoustic feature values including Mel frequency cepstral coefficients (MFCC), spectral centroid, and zero-crossing rate, and sends the acoustic feature values to the scene classification module.
[0018] The feature extraction module includes a Fast Fourier Transform (FFT) calculation stage, a Mel filter bank mapping stage, a Discrete Cosine Transform (DCT) calculation stage, and a parallel computing stage, such as... Figure 2 : Fast Fourier Transform (FFT) calculation stage: Performs fixed-point Fast Fourier Transform (FFT) operations on the preprocessed data and calculates the amplitude spectrum and power spectrum in real time; Mel filter group mapping stage: The amplitude spectrum is weighted and summed using the Mel filter weight matrix to obtain the energy value; Discrete Cosine Transform (DCT) Calculation Stage: The energy values are processed using the Discrete Cosine Transform (DCT) algorithm to obtain the transformed eigenvalues. Parallel computing stage: Three digital signal processor (DSP) parallel cores are started simultaneously to calculate acoustic characteristic values, including Mel frequency cepstral coefficients (MFCC), zero-crossing rate, and spectral centroid, and write them back to shared memory synchronously in frame order.
[0019] During the parallel computing phase, acoustic eigenvalues including Mel-frequency cepstral coefficients (MFCC), zero-crossing rate, and spectral centroid are calculated. (1) Mel frequency cepstral coefficient MFCC sub-stage: The first L terms are extracted from the transformed eigenvalues obtained from the Discrete Cosine Transform (DCT) stage as Mel frequency cepstral coefficients MFCC; (2) Zero-crossing rate sub-stage: The zero-crossing rate is obtained by rapidly accumulating the symbol change count of the preprocessed data frame by frame; (3) Spectral centroid stage: The spectral centroid is obtained by weighted summation of amplitude spectrum and frequency index.
[0020] The feature extraction module performs a Fast Fourier Transform on the preprocessed data to obtain frequency domain information. The power spectrum is then mapped onto a pre-defined Mel filter band. After obtaining the Mel energy, its logarithm is taken, followed by a Discrete Cosine Transform to generate Mel frequency cepstral coefficients (MFCCs). The zero-crossing rate and spectral centroid of each frame are calculated in parallel, providing multi-dimensional feature support for subsequent scene discrimination.
[0021] Scene classification module: Input acoustic feature values into a lightweight decision tree model to achieve scene recognition and output scene labels.
[0022] In the scene classification module, acoustic feature values are input into a lightweight decision tree model to achieve scene recognition and output scene labels. Specifically, as follows: Figure 3 : Step 1: Calculate the low-frequency energy ratio η based on the power spectrum obtained from the Fast Fourier Transform (FFT) calculation stage. If the low-frequency energy ratio η is less than the first set threshold, the environment is determined to be silent; otherwise, proceed to the next step 2. Step 2: Compare the zero-crossing rate ZCR with the second set threshold: If the zero-crossing rate ZCR is less than the second set threshold, it is determined to be a speech environment; otherwise, it is a noise environment, and proceed to the next step 3. Step 3: Compare the spectral centroid with the third set threshold. If the spectral centroid is higher than the third set threshold, it is determined to be a traffic noise environment; if the spectral centroid is not higher than the third set threshold, it is determined to be an indoor noise environment.
[0023] The scene classification module takes MFCC, zero-crossing rate, and spectral centroid as inputs and uses a three-level decision tree structure to classify the environment. It compares each feature with a set threshold to progressively determine the current environment category, improving the accuracy of noise type differentiation. Compared to complex deep learning-based models, the lightweight decision tree algorithm requires less floating-point arithmetic, has a simpler structure, lower computational cost, and is more suitable for embedded DSP platforms. It features fast classification speed, low overall memory overhead, and low inference latency. The first threshold can be set to 0.05, the second threshold to 0.1, and the third threshold to 2kHz.
[0024] Dynamic equalizer configuration module: Adjusts the equalizer's filtering parameters, including filter gain and quality factor Q, in real time based on scene tags, and dynamically configures the number of enabled segments for hardware and software filters to achieve seamless audio effect adjustment.
[0025] The dynamic equalizer configuration module dynamically adjusts equalization parameters based on scene labels. For example, it controls a total of 16 EQ bands, of which 8 are hardware equalizers and the other 8 are software equalizers. The hardware equalizer achieves gain control through audio front-end circuitry, while the software equalizer uses multi-stage IIR filters. Adjustments require setting the activation status and gain value of each band according to different scene types. In quiet environments, to reduce processing load, all equalizer bands except the hardware pass-through channel can be disabled, leaving only the linear channel output. In noisy environments, multiple filters are activated to boost the mid-to-low frequency bands relevant to speech and attenuate the high frequency bands to improve speech clarity. For each activated filter band, the hardware equalizer controls its gain status through configuration registers, while the software equalizer calculates new filter coefficients in the background according to scene requirements. To avoid abrupt noise during parameter switching, the software equalizer uses an atomic switching mechanism for coefficient updates: new coefficients are written all at once after calculation and locked during the update process to ensure a smooth transition of output parameters. After scene recognition and adaptive adjustment, the output of the dynamic equalizer configuration module is the enable status of each channel and the corresponding gain and quality factor Q settings, thereby achieving a dynamic audio equalization effect. This design fully utilizes the advantages of hardware pass-through and segmented filtering, enabling the system to adaptively adjust the frequency response in different environments, significantly reducing the DSP computational burden and power consumption while preserving key sound quality. The combined hardware and software EQ strategy can complete parameter switching within hundreds of microseconds, ensuring real-time performance.
[0026] This scene-aware intelligent audio equalizer adaptive adjustment system achieves a complete process including audio acquisition, multi-dimensional feature extraction, rapid scene classification, dynamic equalizer adjustment, and low-power scheduling management. Through audio feature extraction, decision tree classification algorithms, and dynamic equalizer adjustment, the system can improve the environmental adaptability, real-time performance, and power efficiency of audio processing, reduce computational overhead and energy consumption, identify different audio environments, and adaptively adjust device sound effects according to different situations. It can also fully consider the real-time requirements and energy efficiency needs of embedded systems, providing a highly efficient and flexible audio adaptive processing solution for application scenarios such as mobile devices and in-vehicle systems, helping audio processing technology develop towards intelligence.
[0027] This scene-aware intelligent audio equalizer adaptive adjustment system extracts various features of the audio signal through techniques such as window framing, spectrum analysis, Mel filtering, Discrete Cosine Transform (DCT) conversion, and multi-dimensional feature calculation. While ensuring accuracy, it can effectively reduce audio processing latency and is suitable for embedded platforms with limited hardware performance and resources. The feature extraction module outputs multiple features, including the MFCC coefficient vector, zero-crossing rate, and spectral centroid of each frame. Compared with traditional single feature calculation methods, it significantly reduces computational latency and resource consumption while maintaining the accuracy of audio features through parallel computing and other means.
[0028] The lightweight decision tree model employed in this invention, with its low inference latency and small model size, can accurately determine the current environment category and dynamically respond to changes in environmental noise. It also triggers the corresponding EQ strategy in the dynamic equalizer configuration module, automatically adjusting the activation status, gain, and quality factor Q of each EQ band under different scenarios, thereby effectively improving the clarity and intelligibility of speech and audio. Compared to models based on complex deep learning, the lightweight decision tree has a simple structure, low computational cost, and almost no reliance on large amounts of floating-point operations, making it suitable for running on embedded DSP platforms. It offers advantages such as fast classification speed, low storage consumption, and low inference latency.
[0029] This invention is an edge-based, embedded platform-oriented audio scene recognition and adjustment system utilizing decision trees. It employs audio feature extraction, decision tree classification algorithms, and dynamic EQ adjustment to improve audio processing performance, real-time performance, and power efficiency. This solution aims to significantly reduce computational complexity and power consumption, and shorten end-to-end processing latency while ensuring recognition accuracy and sound quality improvement. This is crucial for meeting the needs of audiophiles and enhancing the user experience, addressing the shortcomings of existing technologies in environmental adaptability, real-time performance, and power efficiency. This invention can perform audio signal sampling and scene recognition, as well as adaptively adjust audio effects. It can be used to identify various types of environmental noise and adjust audio output effects in audio processing scenarios.
[0030] This scene-aware intelligent audio equalizer adaptive adjustment system acquires environmental data through a microphone, and after recognition and processing by algorithms and the system, adaptively adjusts the EQ effect. It is suitable for various audio scenarios such as voice interaction and environmental recognition. By intelligently recognizing audio features and intelligently adjusting audio effect processing, it can make the audio system more intelligent and universal, meeting users' pursuit of audio device sound quality and sound effects.
[0031] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any transformations or substitutions that can be conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A scene-aware intelligent audio equalizer adaptive adjustment system, characterized in that: It includes an audio acquisition and preprocessing module, a feature extraction module, a scene classification module, and a dynamic equalizer configuration module. Audio Acquisition and Preprocessing Module: Uses a microphone to sample and acquire raw audio data in real time, and performs pre-emphasis, frame segmentation, and Hamming windowing operations on the acquired raw audio data to obtain preprocessed data; Feature extraction module: Performs frequency domain transformation on the preprocessed data, calculates acoustic feature values including Mel frequency cepstral coefficients (MFCC), spectral centroid, and zero-crossing rate, and sends the acoustic feature values to the scene classification module; Scene classification module: Inputs acoustic feature values into a lightweight decision tree model to achieve scene recognition and output scene labels; Dynamic equalizer configuration module: Adjusts the equalizer's filtering parameters, including filter gain and quality factor Q, in real time based on scene tags, and dynamically configures the number of enabled segments for hardware and software filters to achieve seamless audio effect adjustment.
2. The scene-aware intelligent audio equalizer adaptive adjustment system as described in claim 1, characterized in that: The audio acquisition and preprocessing module includes a real-time sampling stage, a pre-emphasis stage, a framing stage, and a windowing stage, among which: Real-time sampling stage: The raw audio data acquired by the analog-to-digital converter (ADC) is directly written to the circular buffer at a sampling rate Fs through the direct memory access DMA peripheral. Combined with the timestamp synchronization mechanism, it ensures that no frames are lost. A high-priority interrupt is triggered when the buffer is half full, thus completing jitter-free real-time sampling. Pre-emphasis stage: Perform first-order pre-emphasis filtering; In the framing stage: a dual-buffering alternation strategy based on fixed frame length and frame shift is adopted, combined with the Ping-Pong Buffer mechanism, to automatically switch the read and write areas when each frame of data arrives. Windowing stage: Apply a Hamming window to each frame of audio data to obtain windowed data and store it in a double buffer.
3. The scene-aware intelligent audio equalizer adaptive adjustment system as described in claim 1, characterized in that: The audio acquisition and preprocessing module uses DMA circular buffer technology to achieve zero-copy transmission and is scheduled as a high-priority task in the system.
4. The scene-aware intelligent audio equalizer adaptive adjustment system as described in any one of claims 1-3, characterized in that: The feature extraction module includes a Fast Fourier Transform (FFT) calculation stage, a Mel filter bank mapping stage, a Discrete Cosine Transform (DCT) calculation stage, and a parallel computing stage, among which: Fast Fourier Transform (FFT) calculation stage: Performs fixed-point Fast Fourier Transform (FFT) operations on the preprocessed data and calculates the amplitude spectrum and power spectrum in real time; Mel filter group mapping stage: The amplitude spectrum is weighted and summed using the Mel filter weight matrix to obtain the energy value; Discrete Cosine Transform (DCT) Calculation Stage: The energy values are processed using the Discrete Cosine Transform (DCT) algorithm to obtain the transformed eigenvalues. Parallel computing stage: Three digital signal processor (DSP) parallel cores are started simultaneously to calculate acoustic characteristic values, including Mel frequency cepstral coefficients (MFCC), zero-crossing rate, and spectral centroid, and write them back to shared memory synchronously in frame order.
5. The scene-aware intelligent audio equalizer adaptive adjustment system as described in claim 4, characterized in that: During the parallel computing phase, acoustic eigenvalues including Mel-frequency cepstral coefficients (MFCC), zero-crossing rate, and spectral centroid are calculated. (1) Mel frequency cepstral coefficient MFCC sub-stage: The first L terms are extracted from the transformed eigenvalues obtained from the Discrete Cosine Transform (DCT) stage as Mel frequency cepstral coefficients MFCC; (2) Zero-crossing rate sub-stage: The zero-crossing rate is obtained by rapidly accumulating the symbol change count of the preprocessed data frame by frame; (3) Spectral centroid stage: The spectral centroid is obtained by weighted summation of amplitude spectrum and frequency index.
6. The scene-aware intelligent audio equalizer adaptive adjustment system as described in claim 4, characterized in that: In the scene classification module, acoustic feature values are input into a lightweight decision tree model to achieve scene recognition and output scene labels, specifically: Step 1: Calculate the low-frequency energy ratio η based on the power spectrum obtained from the Fast Fourier Transform (FFT) calculation stage. If the low-frequency energy ratio η is less than the first set threshold, the environment is determined to be silent; otherwise, proceed to the next step 2. Step 2: Compare the zero-crossing rate ZCR with the second set threshold: If the zero-crossing rate ZCR is less than the second set threshold, it is determined to be a speech environment; otherwise, it is a noise environment, and proceed to the next step 3. Step 3: Compare the spectral centroid with the third set threshold. If the spectral centroid is higher than the third set threshold, it is determined to be a traffic noise environment; if the spectral centroid is not higher than the third set threshold, it is determined to be an indoor noise environment.