Balanced adjustment method and system for sound console volume controller
By decoupling volume and frequency control through a dual-channel parallel computing architecture and a deep learning network, the problem of volume and frequency coupling in a mixing console is solved, achieving efficient audio processing in an adaptive noise environment and meeting real-time requirements.
Patent Information
- Application Number
- CN202511897196.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-16
- Publication Date
- 2026-03-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The volume control and frequency equalization of existing mixing consoles are highly coupled, making it difficult to adjust independently and precisely. They also lack environmental adaptability, have large processing delays, and cannot meet real-time requirements.
It adopts a dual-channel parallel computing architecture, combined with deep learning networks and intelligent optimization algorithms, to achieve decoupled control of volume and frequency, adapt to noise environment, and use hardware acceleration technology to improve processing speed.
It enables independent and precise adjustment of volume and frequency, automatically adapts to complex noise environments, reduces the skill requirements for operators, and improves audio quality consistency and real-time performance.
Smart Images
Figure CN121686983A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to mixing console control, and more particularly to a method and system for equalization adjustment of a mixing console volume controller. Background Technology
[0002] As the core device of an audio processing system, the volume control and frequency equalization performance of a mixing console directly determine the final audio output quality. Currently, volume control in mixing consoles is mostly based on analog circuits or digital signal processing (DSP) technology, but several long-standing technical bottlenecks remain. There is a high degree of coupling between volume adjustment and frequency equalization control in traditional mixing consoles. Adjusting the volume often leads to unexpected changes in frequency response characteristics, and vice versa. This mutual interference makes it difficult for operators to independently and precisely adjust volume and frequency; multiple parameters must be repeatedly and collaboratively adjusted to achieve the desired effect, resulting in low operational efficiency.
[0003] Existing systems generally lack the ability to adapt to environmental noise. In complex acoustic environments, such as live performances or noisy conferences, fixed control parameters are insufficient to cope with dynamically changing noise interference, often leading to decreased output audio clarity or a deteriorated signal-to-noise ratio. Optimization of system parameters heavily relies on the operator's auditory experience and subjective judgment, lacking objective and intelligent optimization mechanisms. This not only places high demands on the operator's skills but also makes it difficult to guarantee the consistency and optimality of audio processing results. Traditional audio processing architectures often employ a serial computing model, sequentially completing steps such as signal acquisition, volume control, and frequency equalization. This serial processing method suffers from significant processing delays when dealing with multi-channel, high-sampling-rate audio signals, making it difficult to meet the stringent real-time requirements of modern applications. Summary of the Invention
[0004] Purpose of the invention: The purpose of this invention is to provide a mixing console control method and system that can achieve decoupled control of volume and frequency, has environmental adaptability, can intelligently optimize parameters, and meets high real-time requirements.
[0005] Technical solution: A method for equalization adjustment of a mixing console volume controller, comprising the following steps:
[0006] S1. Acquire audio signals and ambient noise signals from the input of the mixing console, and preprocess the audio signals to obtain an initial system parameter matrix;
[0007] S2. Input the initial system parameter matrix into the deep learning network for inference, extract audio features and predict audio state, and output the audio state data matrix and noise feature data matrix.
[0008] S3. Based on the audio state data matrix and noise feature data matrix, collaborative control is performed through a dual-channel parallel computing architecture. The volume control channel uses first-order active disturbance rejection control to generate the volume control data matrix, and the frequency equalization channel uses second-order active disturbance rejection control to generate the frequency equalization control data matrix.
[0009] S4. Use intelligent optimization algorithms to iteratively optimize the control parameters of the volume control channel and the frequency equalization channel to obtain the optimal control parameter matrix;
[0010] S5. Adjust the transfer functions of the volume control channel and the frequency equalization channel according to the optimal control parameter matrix, and synthesize the output audio signal through the dual-channel parallel computing unit.
[0011] Furthermore, the deep learning network in step S2 is a hybrid architecture of temporal convolutional neural network and long short-term memory network, and an attention mechanism is introduced to focus on the key features of the audio signal.
[0012] Furthermore, in the dual-channel parallel computing architecture in step S3, the volume control channel and the frequency equalization channel interact with each other through a collaborative control unit. The collaborative control unit dynamically adjusts the coupling relationship between the channels based on the audio state data matrix output by the deep learning network.
[0013] Furthermore, the intelligent optimization algorithm in step S4 is an improved particle swarm optimization algorithm, which includes adaptive inertia weight adjustment, chaotic perturbation strategy and phased optimization strategy. The phased optimization strategy includes three stages: global coarse search, local fine search and online fine-tuning.
[0014] An equalization adjustment system for a mixing console volume controller includes:
[0015] The audio acquisition module is used to acquire audio signals from the input of the mixing console and ambient noise signals;
[0016] The preprocessing module is used to preprocess the audio signal to generate an initial system parameter matrix;
[0017] The deep learning inference module is used to extract audio features and predict states based on the initial system parameter matrix, and output an audio state data matrix and a noise feature data matrix.
[0018] A dual-channel parallel control module includes a volume control channel and a frequency equalization channel. The volume control channel is configured as a first-order active disturbance rejection control unit, and the frequency equalization channel is configured as a second-order active disturbance rejection control unit. It is used to generate a control data matrix based on the audio state data matrix and the noise feature data matrix.
[0019] The parameter optimization module is used to iteratively optimize the control parameters through intelligent optimization algorithms and output the optimal control parameter matrix.
[0020] The output synthesis module is used to adjust the channel transfer function according to the optimal control parameter matrix and synthesize the output audio signal.
[0021] Furthermore, the system also includes a hardware acceleration unit, which is a GPU or FPGA, used to implement parallel computing of the dual-channel parallel control module and the deep learning inference module.
[0022] Furthermore, the deep learning inference module integrates a temporal convolutional neural network, a long short-term memory network, and an attention mechanism unit, which is used to calculate audio feature weights.
[0023] Furthermore, the parameter optimization module is configured to execute an improved particle swarm optimization algorithm and includes an adaptive weight adjustment unit, a chaotic perturbation injection unit, and a staged optimization controller to achieve rapid convergence of parameter optimization and search for the global optimum.
[0024] Beneficial effects:
[0025] (1) This invention decouples volume control and frequency equalization into two relatively independent processing channels through an innovative dual-channel parallel computing architecture, and uses a deep learning network for collaborative control, thereby effectively overcoming the problem of mutual interference between the two in traditional technologies, enabling operators to make more precise and independent adjustments to volume and frequency.
[0026] (2) By introducing a deep learning network to perceive and analyze environmental noise characteristics in real time and construct an intelligent mapping relationship between noise and control parameters, this invention enables the system to automatically adapt to different noise environments and dynamically adjust the control strategy, thereby maintaining excellent audio output quality in complex acoustic scenarios.
[0027] (3) This invention employs an improved intelligent optimization algorithm to automatically search for optimal control parameters, eliminating reliance on human experience. This not only reduces the technical requirements for operators but also ensures that the system always operates in a near-optimal state, improving the objective quality and consistency of audio processing.
[0028] (4) Based on the dual-channel parallel computing architecture and supplemented by hardware acceleration technology, this invention changes the traditional serial processing flow to parallel processing, which significantly shortens the total delay of the signal processing link and can well meet the application scenarios with extremely high real-time requirements such as live broadcasts and live performances.
[0029] (5) Thanks to the accurate modeling of audio features by deep learning, the rapid response and strong robustness of the self-disturbance control technology, and the continuous optimization of intelligent algorithms, this invention can output high-quality audio with low distortion and high clarity. Attached Figure Description
[0030] Figure 1 This is a flowchart illustrating the method.
[0031] Figure 2 This is a schematic diagram of the system architecture. Detailed Implementation
[0032] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0033] Example 1:
[0034] This embodiment is applied in a professional recording studio environment. The input audio signal comes from a condenser microphone with a signal level of approximately -60 dBu, and the ambient noise level is below 30 dB SPL. Extremely high sound quality is required for the output audio, with a signal-to-noise ratio greater than 60 dB. The system uses an ARM Cortex-M7 microcontroller and an FPGA hardware acceleration unit.
[0035] The specific implementation process is as follows:
[0036] Step 1: Audio signal acquisition and preprocessing
[0037] Audio Acquisition: A microphone array acquires the audio signal x(t) from the mixing console input and the ambient noise signal n(t). The sampling rate is set to 192kHz and the quantization bit depth is 24 bits.
[0038] Preprocessing: Time-domain sampling of x(t) yields the sequence {x[i]}, i=1,...,N (N=1024); Fast Fourier Transform (FFT) is performed on {x[i]} to obtain the amplitude spectrum A[f] and phase spectrum φ[f], with a frequency resolution of 187.5Hz (192kHz / 1024); Noise features are extracted from n(t): Root mean square (RMS) and spectral flatness are calculated; Initial system parameter matrix M is generated. oo The dimension is 2×512 (corresponding to the characteristics of amplitude spectrum and phase spectrum).
[0039] Assume an audio signal is acquired with an RMS value of 0.5 for x(t) and an RMS value of 0.01 for the ambient noise n(t). After preprocessing, the amplitude spectrum A[f] has a value of 0.8 at 1kHz, and the phase spectrum φ[f] has a value of 0.1rad. Then M oo Some of the elements are:
[0040] M oo=[A[1],A[2],...,A
[512] ;φ[1],φ[2],...,φ
[512] ]=[0.8,...,;0.1,...].
[0041] Step 2: Deep Learning Network Inference
[0042] Network architecture: A hybrid TCNN-LSTM network with attention mechanism is adopted. The TCNN layer uses 1D convolutional kernels (size=3, number of channels=64), the LSTM layer has 128 hidden units, and the attention mechanism calculates feature weights.
[0043] Loss function: , where λ1=0.6, λ2=0.3, λ3=0.1; L reconstruction For the reconstruction loss, mean squared error (MSE) is used. Assume the MSE between the predicted output and the true value is 0.05; L noise For noise robustness loss, based on noise characteristics MSE=0.02; L smoothness For smoothness loss, based on the output rate of change Calculate the loss: .
[0044] Output: The network outputs an audio state data matrix S (128×1 dimension) and a noise feature data matrix N (64×1 dimension). S contains audio feature vectors, such as spectral centroid and zero-crossing rate; N contains the noise type classification results (steady-state noise in this example).
[0045] Step 3: Dual-channel parallel control calculation
[0046] Volume control channel (first-order ADRC):
[0047] Tracking differentiator: v1 = f1(S, ω1) = ω1 × (S - v1), where ω1 = 100 rad / s (bandwidth parameter). Assuming the current value of S is 0.9 and the previous value of v1 is 0.8, then v1 = 100 × (0.9 - 0.8) = 10 (differential output).
[0048] State observer: x1=f2(v1,b1)=b1×v1, where b1=0.5 (control gain). X1=0.5×10=5.
[0049] Control quantity: u1=f3(x1,S)=x1+k1×S, k1=0.1. U1=5+0.×0.9=5.09.
[0050] Output volume control data matrix U1=[u1]=[5.09].
[0051] Frequency equalization channel (second-order ADRC):
[0052] Tracking differentiator: v2=f4(U1,ω2)=ω2×(U1-v2), where ω2=50rad / s. U1=5.09, the previous value of v2 is 5.0, then v2=50×(5.09-5.0)=4.5.
[0053] Extended state observer: x2=f5(v2,b2,N)=b2×v2+k2×N, where b2=0.3, k2=0.05, N=0.1 (noise characteristic). X2=0.3×4.5+0.05×0.1=1.35+0.005=1.355.
[0054] Control quantity: , , . .
[0055] Output frequency equalization control data matrix U2=[u2]=[2.374].
[0056] Cooperative control unit: The coupling matrix is dynamically adjusted based on the basis S and N. In this example, the coupling matrix C = [0.9, 0.1; 0.1, 0.9] ensures that the two channels are independent yet cooperative.
[0057] Step 4: Intelligent Parameter Optimization (Improved PSO Algorithm)
[0058] Algorithm parameters: Number of particles = 20, maximum number of iterations (max) iter =100, adaptive inertia weight w max =0.9, w min =0.4, acceleration constants c1=2.0, c2=2.0, chaotic perturbation Δchaos=0.01×randn().
[0059] Fitness function: F = α × sound quality index + β × stability index + γ × real-time performance index, where α = 0.5, β = 0.3, and γ = 0.2. Assume the current particle position is x. i (k)=[0.5,0.6], individual optimality pbest i =[0.5,0.6], global optimal gbest=[0.5,0.6]; sound quality index=0.8 (based on signal-to-noise ratio), stability index=0.9, real-time index=0.95 (latency <5ms). F=0.5×0.8+0.3×0.9+0.2×0.95=0.4+0.27+0.19=0.86.
[0060] PSO Update:
[0061] Current iteration Inertial weight .
[0062] Speed update: v i (k+1)=w×v i (k)+c1×r1×(pbest i -x i (k))+c2×r2×(gbest-x i (k))+Δchaos. Assume , , , v i (k+1)=0.75×[0.1,0.1]+2.0×0.5×([0.5,0.6]-[0.5,0.6])+2.0×0.6×([0 .5,0.6]-[0.5,0.6])+0.003=[0.075,0.075]+0+0+0.003=[0.078,0.078].
[0063] Location update: , x i (k+1)=[0.5,0.6]+[0.078,0.078]×0.1=[0.5,0.6]+[0.0078,0.0078]=[0.5078,0.6078].
[0064] Phased optimization: k=30 belongs to phase 2 (local search), with moderate inertia weight. Output the optimal control parameter matrix. .
[0065] Step 5: Dual-channel parallel computing and output
[0066] Adjust the transfer functions: Volume control channel transfer function G1(s)=P[1] / (s+ω1)=0.5078 / (s+100), frequency equalization channel transfer function G2(s)=P[2] / (s 2 +2ζω2s+ω2 2 )=0.6078 / (s 2 +70.7s+2500) (ζ=0.707).
[0067] Synthesized output: Y = G1(s) × U1 + G2(s) × U2. Discrete calculation in the time domain is performed using a bilinear transform. Assuming that after discretization, the output signal y[n] = 0.9 × u1 + 0.8 × u2 = 0.9 × 5.09 + 0.8 × 2.374 = 4.581 + 1.899 = 6.48.
[0068] D / A conversion: Outputs analog audio signal with a level of approximately -10dBu and a signal-to-noise ratio of >60dB.
[0069] The system architecture used in this embodiment is as follows:
[0070] Hardware: ARM Cortex-M7 microcontroller handles control logic, FPGA implements TCNN-LSTM network and parallel computation of dual-channel ADRC.
[0071] Software: FreeRTOS real-time operating system, TensorFlow Lite for deep learning inference, and FFTW library for spectrum analysis.
[0072] Example 2:
[0073] This embodiment is applied to a live streaming scenario. The input audio signal comes from a line input with a level of approximately -10 dBu, and the ambient noise level is 40-60 dB SPL. High real-time performance is required, with a latency of less than 10 ms. The system uses an NVIDIA Jetson GPU as the hardware acceleration unit.
[0074] The specific implementation process is as follows:
[0075] Step 1: Audio signal acquisition and preprocessing
[0076] Audio acquisition: x(t) is acquired from the line input, and n(t) is acquired from the microphone array. The sampling rate is 48kHz, and the frame length is 512 points.
[0077] Preprocessing: FFT was used to obtain the amplitude and phase spectra with a frequency resolution of 93.75 Hz; Noise feature extraction: Noise spectral entropy and zero-crossing rate were calculated; Initial system parameter matrix M was obtained. oo Dimension 2×256. Assume the RMS value of x(t) is 0.3, and the RMS value of n(t) is 0.05. The amplitude spectrum is 0.6 at 1 kHz, and the phase spectrum is 0.2 rad. M oo =[0.6,...;0.2,...].
[0078] Step 2: Deep Learning Network Inference
[0079] Network inference: The TCNN-LSTM network has an attention mechanism, and the loss function parameters are λ1=0.5, λ2=0.4, and λ3=0.1. , , . L=0.5×0.04+0.4×0.03+0.1×0.02=0.02+0.012+0.002=0.034.
[0080] Output: S (dimension 64×1), N (dimension 32×1), noise type is non-steady-state noise.
[0081] Step 3: Dual-channel parallel control calculation
[0082] Volume control channel: ω1=150rad / s, b1=0.6; S=0.7, previous value of v1=0.65→v1=150×(0.7-0.65)=7.5; x1=0.6×7.5=4.5; u1=4.5+0.1×0.7=4.57; U1=[4.57].
[0083] Frequency equalization channel: ω2=80rad / s, b2=0.4, k2=0.1, N=0.2; v2=80×(4.57-4.5)=5.6; x2=0.4×5.6+0.1×0.2=2.24+0.02=2.26; u2=2.26+0.2×4.57+0.01×0.2=2.26+0.914+0.002=3.176; U2=[3.176].
[0084] Cooperative control: coupling matrix C = [0.8, 0.2; 0.2, 0.8].
[0085] Step 4: Intelligent Parameter Optimization (Improved PSO Algorithm)
[0086] Parameter: max iter =50, w max =0.9, w min =0.3, c1=1.8, c2=2.2.
[0087] Fitness function: α=0.6, β=0.2, γ=0.2. Sound quality index = 0.7, stability index = 0.8, real-time performance index = 0.9 (latency < 8ms). F = 0.6 × 0.7 + 0.2 × 0.8 + 0.2 × 0.9 = 0.42 + 0.16 + 0.18 = 0.76.
[0088] PSO Update:
[0089] k=10, w=0.9-(0.9-0.3)×10 / 50=0.9-0.12=0.78.
[0090] x i (k)=[0.4,0.5], v i (k)=[0.05,0.05], pbest i =[0.4,0.5], gbest=[0.4,0.5], r1=0.4, r2=0.7, Δchaos=0.01×0.2=0.002.
[0091] v i(k+1)=0.78×[0.05,0.05]+1.8×0.4×([0.4,0.5]-[0.4,0.5])+2.2×0.7×([ 0.4,0.5]-[0.4,0.5])+0.002=[0.039,0.039]+0+0+0.002=[0.041,0.041].
[0092] x i (k+1)=[0.4,0.5]+[0.041,0.041]×0.1=[0.4041,0.5041].
[0093] Phased optimization: k=10 belongs to phase 1 (global search), with a large inertia weight. .
[0094] Step 5: Dual-channel parallel computing and output
[0095] Transfer functions: G1(s) = 0.4041 / (s+150), G2(s) = 0.5041 / (s+150) 2 +113.1s+6400) (ζ=0.707).
[0096] Output composition: y[n]=0.85×u1+0.75×u2=0.85×4.57+0.75×3.176=3.8845+2.382=6.2665.
[0097] D / A conversion: outputs audio signal with a delay controlled within 8ms.
[0098] The system architecture used in this embodiment is as follows:
[0099] Hardware: NVIDIA Jetson GPUs handle deep learning and parallel computing, while microcontrollers manage I / O.
[0100] Software: Android application interface, PyTorch Mobile for deep learning.
[0101] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.
Claims
1. A method of equalizing the fader control of a sound console, characterized by, The method comprises the following steps: S1, collecting audio signals at the input end of a mixing console and environmental noise signals, and pre-processing the audio signals to obtain an initial system parameter matrix; S2, inputting the initial system parameter matrix into a deep learning network for inference, extracting audio features and predicting audio states, and outputting an audio state data matrix and a noise feature data matrix; S3, based on the audio state data matrix and the noise feature data matrix, performing collaborative control through a double-channel parallel computing architecture, wherein a volume control channel adopts first-order active disturbance rejection control to generate a volume control data matrix, and a frequency equalization channel adopts second-order active disturbance rejection control to generate a frequency equalization control data matrix; S4, using an intelligent optimization algorithm to iteratively optimize the control parameters of the volume control channel and the frequency equalization channel to obtain an optimal control parameter matrix; S5, adjusting the transfer functions of the volume control channel and the frequency equalization channel according to the optimal control parameter matrix, and synthesizing and outputting an audio signal through a double-channel parallel computing unit.
2. A method of equalization adjustment of a fader of a sound console according to claim 1, characterized by, The deep learning network in step S2 is a hybrid architecture of a time convolution neural network and a long short-term memory network, and an attention mechanism is introduced to focus on key features of the audio signal.
3. The method of equalization adjustment of a fader of a sound console according to claim 1, wherein, In the double-channel parallel computing architecture in step S3, the volume control channel and the frequency equalization channel interact through a collaborative control unit, which dynamically adjusts the coupling relationship between the channels based on the audio state data matrix output by the deep learning network.
4. The method of equalization adjustment of a fader of a sound console according to claim 1, wherein, The intelligent optimization algorithm in step S4 is an improved particle swarm optimization algorithm, which includes adaptive inertia weight adjustment, chaos disturbance strategy and stage optimization strategy, wherein the stage optimization strategy includes global coarse search, local fine search and online fine tuning.
5. A system for equalizing the faders of a mixing console for implementing the method of any one of claims 1 to 4, characterized by It comprises: An audio acquisition module for acquiring audio signals at the input end of a mixing console and environmental noise signals; A preprocessing module for pre-processing the audio signals to generate an initial system parameter matrix; A deep learning inference module for audio feature extraction and state prediction based on the initial system parameter matrix, outputting an audio state data matrix and a noise feature data matrix; A double-channel parallel control module, including a volume control channel and a frequency equalization channel, the volume control channel being configured as a first-order active disturbance rejection control unit, and the frequency equalization channel being configured as a second-order active disturbance rejection control unit, for generating a control data matrix according to the audio state data matrix and the noise feature data matrix; A parameter optimization module for iteratively optimizing the control parameters through an intelligent optimization algorithm to output an optimal control parameter matrix; An output synthesis module for adjusting the channel transfer functions according to the optimal control parameter matrix and synthesizing an output audio signal.
6. A fader equalization system for a sound console as defined in claim 5, wherein, The system further comprises a hardware acceleration unit, which is a GPU or FPGA, for realizing parallel computing of the double-channel parallel control module and the deep learning inference module.
7. A fader equalization system for a sound console as defined in claim 5, wherein, The deep learning inference module integrates a time convolution neural network, a long short-term memory network and an attention mechanism unit, and the attention mechanism unit is used to calculate the audio feature weight.
8. The equalization adjustment system for a fader of a sound console of claim 5, wherein, The parameter optimization module is configured to execute an improved particle swarm optimization algorithm, and comprises an adaptive weight adjustment unit, a chaotic disturbance injection unit and a staged optimization controller to realize fast convergence and global optimal solution search of parameter optimization.