Pickup and pickup system based on artificial intelligence noise reduction
By adopting artificial intelligence-based audio processing technology in the pickup system, including voiceprint feature extraction, noise classification suppression and deep learning noise reduction, the problem that existing pickups cannot effectively filter environmental noise is solved, and the quality of the audio signal is significantly improved.
Patent Information
- Application Number
- CN202510223093.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-27
- Publication Date
- 2025-06-13
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing pickups cannot effectively filter ambient noise, affecting the quality of the audio signal.
Adopt a pickup system based on artificial intelligence, including an audio acquisition module, an audio processing unit and an output quality monitoring module. The audio processing unit performs noise reduction processing of audio signals through technologies such as voiceprint feature extraction, noise classification suppression, multi-scale fusion enhancement and deep learning noise reduction.
Effectively filter ambient noise, improve the quality of audio signals, and significantly improve the performance of the pickup.
Smart Images
Figure CN120148523A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of audio signal processing, and particularly to a pickup and a pickup system based on artificial intelligence noise reduction. Background Art
[0002] A pickup, as an indispensable key device in the field of audio signal processing, is of great importance. With its precise audio acquisition ability, it plays a crucial role in many scenarios such as meeting recording, monitoring and listening, distance education, speech recognition, live broadcast, and telecommuting. The working principle of traditional pickups is relatively simple. They mainly capture sound signals in the surrounding environment through audio acquisition components such as microphones and convert them into electrical signals for output. These electrical signals are then transmitted to subsequent processing devices, such as audio amplifiers, recording devices, or digital signal processing systems, etc., to achieve amplification, storage, or further analysis of the sound.
[0003] However, existing pickups usually directly output all the collected sound signals and cannot effectively filter out environmental noise, thereby affecting the quality of audio signals. Summary of the Invention
[0004] The purpose of the present invention is to provide a pickup and a pickup system based on artificial intelligence noise reduction, aiming to solve the technical problem that existing pickups usually directly output all the collected sound signals and cannot effectively filter out environmental noise, thereby affecting the quality of audio signals.
[0005] To achieve the above purpose, a pickup based on artificial intelligence noise reduction adopted by the present invention includes an audio acquisition module, a raw audio output port, a classification and transmission module, an audio processing unit, an AI noise reduction human voice output port, an output quality monitoring module, a feedback module, and a model optimization unit; the audio processing unit includes a voiceprint feature extraction module, a noise classification and suppression module, a multi-scale fusion and enhancement module, and a deep learning noise reduction sub-module;
[0006] Both the raw audio output port and the AI noise reduction human voice output port are connected to the classification and transmission module, the classification and transmission module is connected to both the audio processing unit and the audio acquisition module, and the audio processing unit is connected to the audio acquisition module; the output quality monitoring module is connected to the AI noise reduction human voice output port, and the feedback module and the model optimization unit are respectively connected to the output quality monitoring module and the audio processing unit;
[0007] The voiceprint feature extraction module is connected to the audio acquisition module, the noise classification and suppression module is connected to the voiceprint feature extraction module, the multi-scale fusion and enhancement module is connected to the noise classification and suppression module, and the deep learning noise reduction sub-module is connected to the multi-scale fusion and enhancement module;
[0008] The audio acquisition module is used to capture audio signals in the surrounding environment and convert them into digital signals for subsequent processing;
[0009] The original audio output port is connected to the classification and transmission module, and the original audio signal is obtained from the audio acquisition module through the classification and transmission module for output;
[0010] The audio processing unit performs a series of processes on the acquired audio signals, including voiceprint feature extraction, noise classification and suppression, multi-scale fusion enhancement, and deep learning noise reduction;
[0011] The AI noise reduction human voice output port is connected to the classification and transmission module, and the processed audio signal is transmitted to an external device through the classification and transmission module.
[0012] Among them, the model optimization unit includes an acquisition module, an update strategy selection module, a model update execution module, and a model update verification module. The model update execution module is connected to the audio processing unit. The update strategy selection module is arranged between the model update execution module and the acquisition module. The model update verification module is connected to the audio processing unit;
[0013] The acquisition module is used to collect new data samples and perform preprocessing;
[0014] The update strategy selection module selects a suitable update strategy based on the acquired data samples and the model performance evaluation results of the current audio processing unit;
[0015] The model update execution module updates the model according to the update strategy;
[0016] The model update verification module is used to verify the updated audio processing module.
[0017] Among them, the pick-up microphone based on artificial intelligence noise reduction further includes a security protection module, and the security protection module is connected to the audio processing unit;
[0018] The security protection module uses encryption technology, firewalls, and intrusion detection to improve the security and reliability of the pick-up microphone.
[0019] Among them, the pick-up microphone based on artificial intelligence noise reduction further includes an adaptive learning module, and the adaptive learning module is connected to the model optimization unit.
[0020] The present invention also provides a pick-up system based on artificial intelligence noise reduction, including the pick-up microphone based on artificial intelligence noise reduction as described above,
[0021] The sound pickup system based on artificial intelligence noise reduction further includes a user interaction module, which is used for users to configure the working mode and parameters of the sound pickup based on artificial intelligence noise reduction according to their needs through a mobile application, a web page or a local control panel.
[0022] Among them, the sound pickup system based on artificial intelligence noise reduction further includes a configuration module, and the configuration module is connected to the user interaction module.
[0023] Among them, the sound pickup system based on artificial intelligence noise reduction further includes a login module, an identity authentication module and a permission management module. The login module is connected to the user interaction module, the identity authentication module is connected to the login module, and the permission management module is connected to the identity authentication module.
[0024] In the specific use of a sound pickup and a sound pickup system based on artificial intelligence noise reduction according to the present invention, first, the audio acquisition module captures audio signals in the surrounding environment and converts them into digital signals for subsequent processing, and forms two channels of audio signals. One channel of audio signal is output as the original audio signal through the classification and transmission module to the original audio output port, and the other channel of audio signal is transmitted to the audio processing unit. The voiceprint feature extraction module extracts the voiceprint features in the audio signal by using a deep learning model (such as a convolutional neural network, CNN) and conducts feature vector comparison to distinguish human voices and noises; the noise classification and suppression module classifies the noises by using a machine learning algorithm (such as a support vector machine, SVM), and classifies the noises into common types (such as air conditioner noise, keyboard noise, traffic noise, etc.). And according to the classification results, different noise reduction algorithms (such as spectral subtraction, Wiener filtering, etc.) are applied for noise suppression; the multi-scale fusion and enhancement module draws on multi-scale fusion technology (such as the multi-scale receptive field enhancement module in FCM-YOLO), strengthens the correlation between feature information through a multi-branch structure, and uses context attention technology to reduce the influence of background noise on human voices; the deep learning noise reduction sub-module uses a generative adversarial network (GAN) or a variational autoencoder (VAE) to perform noise reduction processing on the audio signal. After the processing is completed, the processed audio signal containing only human voices is output from the AI noise reduction human voice output port after being classified by the classification and transmission module. At the same time, the output quality monitoring module can detect the audio signal output from the AI noise reduction human voice output port and feedback the detection result through the feedback module, thereby solving the technical problem that the existing sound pickups usually directly output all the collected sound signals and cannot effectively filter environmental noises, thus affecting the quality of the audio signals. Description of the Drawings
[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0026] Figure 1 It is the principle block diagram of the pickup based on artificial intelligence noise reduction of the present invention.
[0027] Figure 2 It is the principle block diagram of the pickup system based on artificial intelligence noise reduction of the present invention.
[0028] 101 - Audio acquisition module, 102 - Original audio output port, 103 - Classification and transmission module, 104 - Audio processing unit, 105 - AI noise reduction human voice output port, 106 - Output quality monitoring module, 107 - Feedback module, 108 - Model optimization unit, 109 - Security protection module, 110 - Adaptive learning module, 111 - Voiceprint feature extraction module, 112 - Noise classification and suppression module, 113 - Multi-scale fusion and enhancement module, 114 - Deep learning noise reduction sub-module, 115 - Acquisition module, 116 - Update strategy selection module, 117 - Model update execution module, 118 - Model update verification module, 201 - User interaction module, 202 - Configuration module, 203 - Login module, 204 - Identity authentication module, 205 - Permission management module, 206 - Fault detection module, 207 - Alarm module. Detailed implementation manners
[0029] The following will describe in detail the embodiments of the present invention. The examples of the embodiments are shown in the drawings. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention and should not be construed as a limitation of the present invention.
[0030] Please refer to Figure 1 , Figure 1 It is the principle block diagram of the pickup based on artificial intelligence noise reduction of the present invention.
[0031] The present invention provides a pickup based on artificial intelligence noise reduction, which includes an audio acquisition module 101, a raw audio output port 102, a classification and transmission module 103, an audio processing unit 104, an AI noise reduction human voice output port 105, an output quality monitoring module 106, a feedback module 107, a model optimization unit 108, a security protection module 109, and an adaptive learning module 110; the audio processing unit 104 includes a voiceprint feature extraction module 111, a noise classification and suppression module 112, a multi-scale fusion and enhancement module 113, and a deep learning noise reduction sub-module 114, and the model optimization unit 108 includes an acquisition module 115, an update strategy selection module 116, a model update execution module 117, and a model update verification module 118; the foregoing solution solves the technical problem in the prior art that pickups usually directly output all the collected sound signals, unable to effectively filter environmental noise, thereby affecting the quality of audio signals.
[0032] For this specific embodiment, the audio acquisition module 101 is used to capture audio signals in the surrounding environment and convert them into digital signals for subsequent processing;
[0033] The raw audio output port 102 is connected to the classification and transmission module 103, and the raw audio signal is obtained from the audio acquisition module 101 through the classification and transmission module 103 and output;
[0034] The audio processing unit 104 performs a series of processes on the collected audio signals, including voiceprint feature extraction, noise classification and suppression, multi-scale fusion and enhancement, and deep learning noise reduction;
[0035] The AI noise reduction human voice output port 105 is connected to the classification and transmission module 103, and the processed audio signal is transmitted to an external device through the classification and transmission module 103.
[0036] The voiceprint feature extraction module 111 uses a deep learning model (such as a convolutional neural network, CNN) to extract voiceprint features in the audio signal, and performs feature vector comparison to distinguish human voices and noise;
[0037] The noise classification and suppression module 112 classifies noise using machine learning algorithms (such as support vector machines, SVMs), classifies the noise into common types (such as air conditioner noise, keyboard noise, traffic noise, etc.), and applies different noise reduction algorithms (such as spectral subtraction, Wiener filtering, etc.) for noise suppression according to the classification results;
[0038] The multi-scale fusion and enhancement module 113 draws on multi-scale fusion technologies (such as the multi-scale receptive field enhancement module in FCM-YOLO), strengthens the correlation between feature information through a multi-branch structure, and uses context attention technology to reduce the impact of background noise on human voices;
[0039] The deep learning noise reduction sub-module 114 uses a generative adversarial network (GAN) or a variational autoencoder (VAE) to perform noise reduction processing on the audio signal.
[0040] Among them, the original audio output port 102 and the AI noise reduction human voice output port 105 are both connected to the classification and transmission module 103. The classification and transmission module 103 is connected to the audio processing unit 104 and the audio acquisition module 101, and the audio processing unit 104 is connected to the audio acquisition module 101. The output quality monitoring module 106 is connected to the AI noise reduction human voice output port 105, and the feedback module 107 and the model optimization unit 108 are respectively connected to the output quality monitoring module 106 and the audio processing unit 104;
[0041] The voiceprint feature extraction module 111 is connected to the audio acquisition module 101, the noise classification and suppression module 112 is connected to the voiceprint feature extraction module 111, the multi-scale fusion and enhancement module 113 is connected to the noise classification and suppression module 112, and the deep learning noise reduction sub-module 114 is connected to the multi-scale fusion and enhancement module 113;
[0042] In specific use, first, the audio acquisition module 101 captures audio signals in the surrounding environment, converts them into digital signals for subsequent processing, and forms two channels of audio signals. One channel of the audio signal is output as the original audio signal through the classification and transmission module 103 to the original audio output port 102. The other channel of the audio signal is transmitted to the audio processing unit 104. The voiceprint feature extraction module 111 uses a deep learning model (such as a convolutional neural network, CNN) to extract the voiceprint features in the audio signal, and conducts feature vector comparison to distinguish human voices and noises. The noise classification and suppression module 112 classifies the noises using a machine learning algorithm (such as a support vector machine, SVM), classifies the noises into common types (such as air conditioner noise, keyboard noise, traffic noise, etc.), and applies different noise reduction algorithms (such as spectral subtraction, Wiener filtering, etc.) for noise suppression according to the classification results. The multi-scale fusion and enhancement module 113 draws on multi-scale fusion technology (such as the multi-scale receptive field enhancement module in FCM-YOLO), strengthens the correlation between feature information through a multi-branch structure, and uses context attention technology to reduce the influence of background noise on human voices. The deep learning noise reduction sub-module 114 uses a generative adversarial network (GAN) or a variational autoencoder (VAE) to perform noise reduction processing on the audio signal. After the processing is completed, it is classified by the classification and transmission module 103 and then output as the processed audio signal containing only human voices from the AI noise reduction human voice output port 105. At the same time, the output quality monitoring module 106 can detect the audio signal output from the AI noise reduction human voice output port 105, and feedback the detection result through the feedback module 107, thereby solving the technical problem in the prior art that a pick-up usually directly outputs all the collected sound signals, unable to effectively filter environmental noises, and thus affecting the quality of the audio signal.
[0043] Secondly, the model update execution module 117 is connected to the audio processing unit 104. The update strategy selection module 116 is arranged between the model update execution module 117 and the acquisition module 115. The model update verification module 118 is connected to the audio processing unit 104.
[0044] The acquisition module 115 is used to collect new data samples and perform preprocessing.
[0045] The update strategy selection module 116 selects a suitable update strategy based on the collected data samples and the model performance evaluation result of the current audio processing unit 104.
[0046] The model update execution module 117 updates the model according to the update strategy.
[0047] The model update verification module 118 is used to verify the updated audio processing module.
[0048] First, the acquisition module 115 collects new data samples and preprocesses them to ensure the quality and consistency of the data. Then, the update strategy selection module 116 selects an appropriate update strategy based on the current model performance evaluation results of the audio processing unit 104 and the newly collected data samples. These strategies may include full update, incremental update, or hybrid update, etc. Next, the model update execution module 117 updates the model in the audio processing unit 104 according to the selected strategy. Finally, the model update verification module 118 verifies the updated model to ensure that its performance and stability meet the expectations.
[0049] Meanwhile, the security protection module 109 is connected to the audio processing unit 104;
[0050] The security protection module 109 uses encryption technology, firewalls, and intrusion detection to improve the security and reliability of the pickup.
[0051] In addition, the adaptive learning module 110 is connected to the model optimization unit 108. The adaptive learning module 110 can continuously learn new audio signal features and noise types and feedback this information to the model optimization unit 108.
[0052] The model optimization unit 108 adjusts and optimizes the model according to this information, enabling the model to better adapt to complex and changing environmental noises and audio signal features, and improving the noise reduction effect and audio quality.
[0053] When using a microphone based on artificial intelligence noise reduction in this embodiment, during specific use, first, the audio acquisition module 101 captures audio signals in the surrounding environment and converts them into digital signals for subsequent processing, forming two channels of audio signals. One channel of audio signal is output as the original audio signal through the classification and transmission module 103 to the original audio output port 102. The other channel of audio signal is transmitted to the audio processing unit 104. The voiceprint feature extraction module 111 uses a deep learning model (such as a convolutional neural network, CNN) to extract the voiceprint features in the audio signal and conducts feature vector comparison to distinguish between human voices and noise. The noise classification and suppression module 112 classifies the noise using a machine learning algorithm (such as a support vector machine, SVM), classifying the noise into common types (such as air conditioner noise, keyboard noise, traffic noise, etc.). And according to the classification results, different noise reduction algorithms (such as spectral subtraction, Wiener filtering, etc.) are applied for noise suppression. The multi-scale fusion and enhancement module 113 draws on multi-scale fusion technology (such as the multi-scale receptive field enhancement module in FCM-YOLO), strengthens the correlation between feature information through a multi-branch structure, and uses context attention technology to reduce the influence of background noise on human voices. The deep learning noise reduction sub-module 114 uses a generative adversarial network (GAN) or a variational autoencoder (VAE) to perform noise reduction processing on the audio signal. After processing, it is classified by the classification and transmission module 103 and the processed audio signal containing only human voices is output from the AI noise reduction human voice output port 105. At the same time, the output quality monitoring module 106 can detect the audio signal output from the AI noise reduction human voice output port 105 and feedback the detection result through the feedback module 107. In this way, it solves the technical problem that the microphones in the prior art usually directly output all the collected sound signals and cannot effectively filter environmental noise, thereby affecting the quality of the audio signal.
[0054] Please refer to Figure 2 , Figure 2 which is the principle block diagram of the microphone system based on artificial intelligence noise reduction of the present invention.
[0055] The present invention also provides a microphone system based on artificial intelligence noise reduction, including the microphone based on artificial intelligence noise reduction as described above, and further including a user interaction module 201, a configuration module 202, a login module 203, an identity authentication module 204, a permission management module 205, a fault detection module 206, and an alarm module 207.
[0056] For this specific embodiment, the user interaction module 201 is used for users to configure the working mode and parameters of the microphone based on artificial intelligence noise reduction according to their needs through a mobile application, a web page, or a local control panel.
[0057] Among them, the configuration module 202 is connected to the user interaction module 201, and the configuration module 202 is used to configure the working mode and parameters of the pick-up based on artificial intelligence noise reduction.
[0058] Secondly, the login module 203 is connected to the user interaction module 201, the authentication module 204 is connected to the login module 203, the permission management module 205 is connected to the authentication module 204. The login module 203 is used for the user to log in to the user interaction module 201 and perform verification on the authentication module 204. At the same time, the permission management module 205 assigns different management permissions according to the logged-in user identity.
[0059] Thirdly, the fault detection module 206 is connected to the audio processing unit 104, and the alarm module 207 is connected to the fault diagnosis module. When the audio processing unit 104 fails, the fault detection module 206 can detect it in time and give an alarm through the alarm module 207.
[0060] When using the pick-up based on artificial intelligence noise reduction and the pick-up system of this embodiment, in specific use, the user interaction module 201 is used for the user to configure the working mode and parameters of the pick-up based on artificial intelligence noise reduction according to requirements through a mobile phone application, a web page or a local control panel. The configuration module 202 is used to configure the working mode and parameters of the pick-up based on artificial intelligence noise reduction. The login module 203 is used for the user to log in to the user interaction module 201 and perform verification on the authentication module 204. At the same time, the permission management module 205 assigns different management permissions according to the logged-in user identity. When the audio processing unit 104 fails, the fault detection module 206 can detect it in time and give an alarm through the alarm module 207.
[0061] The above-disclosed is only a preferred embodiment of the present invention. Of course, it cannot be used to limit the scope of the rights of the present invention. Those of ordinary skill in the art can understand the whole or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.
Claims
1. A microphone based on artificial intelligence noise reduction, characterized in that: It includes an audio acquisition module, an original audio output port, a classification transmission module, an audio processing unit, an AI noise reduction human voice output port, an output quality monitoring module, a feedback module and a model optimization unit; the audio processing unit includes a voiceprint feature extraction module, a noise classification suppression module, a multi-scale fusion enhancement module and a deep learning noise reduction submodule; The original audio output port and the AI noise reduction human voice output port are both connected to the classification transmission module, the classification transmission module is both connected to the audio processing unit and the audio acquisition module, and the audio processing unit is connected to the audio acquisition module; the output quality monitoring module is connected to the AI noise reduction human voice output port, and the feedback module and the model optimization unit are respectively connected to the output quality monitoring module and the audio processing unit; The voiceprint feature extraction module is connected to the audio acquisition module, the noise classification suppression module is connected to the voiceprint feature extraction module, the multi-scale fusion enhancement module is connected to the noise classification suppression module, and the deep learning denoising submodule is connected to the multi-scale fusion enhancement module; The audio acquisition module is used to capture audio signals in the surrounding environment and convert them into digital signals for subsequent processing; The original audio output port is connected to the classification transmission module, and the original audio signal output is obtained from the audio acquisition module through the classification transmission module; The audio processing unit performs a series of processing on the collected audio signal, including voiceprint feature extraction, noise classification suppression, multi-scale fusion enhancement and deep learning noise reduction; The AI noise reduction voice output port is connected to the classification transmission module, and the processed audio signal is transmitted to an external device through the classification transmission module.
2. The microphone based on artificial intelligence noise reduction as claimed in claim 1, characterized in that: The model optimization unit includes an acquisition module, an update strategy selection module, a model update execution module and a model update verification module, the model update execution module is connected to the audio processing unit, the update strategy selection module is arranged between the model update execution module and the acquisition module, and the model update verification module is connected to the audio processing unit; The acquisition module is used to collect new data samples and perform preprocessing; The update strategy selection module selects a suitable update strategy based on the collected data samples and the current model performance evaluation result of the audio processing unit; The model update execution module performs model update according to the update strategy; The model update verification module is used to verify the updated audio processing module.
3. The microphone based on artificial intelligence noise reduction as claimed in claim 2, characterized in that: The microphone based on artificial intelligence noise reduction further includes a safety protection module, and the safety protection module is connected to the audio processing unit; The security protection module adopts encryption technology, firewall, and intrusion detection to improve the security and reliability of the microphone.
4. The microphone based on artificial intelligence noise reduction as claimed in claim 3, characterized in that: The microphone based on artificial intelligence noise reduction also includes an adaptive learning module, which is connected to the model optimization unit.
5. The microphone based on artificial intelligence noise reduction as claimed in claim 4, characterized in that: The voiceprint feature extraction module uses a deep learning model to extract voiceprint features from audio signals and compares feature vectors to distinguish between human voice and noise. The noise classification and suppression module classifies the noise using a machine learning algorithm, classifies the noise into common types, and applies different noise reduction algorithms to suppress the noise according to the classification results; The multi-scale fusion enhancement module draws on the multi-scale fusion technology to strengthen the correlation between feature information through a multi-branch structure, and uses contextual attention technology to reduce the impact of background noise on human voice; The deep learning denoising submodule uses a generative adversarial network or a variational autoencoder to perform denoising on the audio signal.
6. A sound pickup system based on artificial intelligence noise reduction, comprising the sound pickup based on artificial intelligence noise reduction as claimed in claim 5, characterized in that: The sound pickup system based on artificial intelligence noise reduction also includes a user interaction module, which is used for the user to configure the working mode and parameters of the sound pickup based on artificial intelligence noise reduction according to needs through a mobile phone application, a web page or a local control panel.
7. The sound pickup system based on artificial intelligence noise reduction as claimed in claim 6, characterized in that: The sound pickup system based on artificial intelligence noise reduction also includes a configuration module, and the configuration module is connected to the user interaction module.
8. The sound pickup system based on artificial intelligence noise reduction as claimed in claim 7, characterized in that: The sound pickup system based on artificial intelligence noise reduction also includes a login module, an identity authentication module and a rights management module. The login module is connected to the user interaction module, the identity authentication module is connected to the login module, and the rights management module is connected to the identity authentication module.