Hearing aid audio classification method
By combining deep learning and traditional audio feature extraction with a hearing aid audio classification method, and utilizing neural network models and attention mechanisms, the problem of insufficient audio signal processing capabilities of hearing aids in complex environments is solved, thereby improving classification accuracy and stability and meeting users' personalized needs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2026-03-13
AI Technical Summary
Traditional hearing aids have limited ability to recognize and process audio signals in complex environments, making them difficult to adapt to various environments and affecting their effectiveness.
The method for audio classification of hearing aids includes data acquisition and preprocessing, construction of neural network models, real-time audio input, scene recognition and personalized adjustment. It combines deep learning and traditional audio feature extraction, classifies audio signals through convolutional neural networks and recurrent neural networks, and optimizes the model using attention mechanisms and transfer learning.
It improves the accuracy and stability of audio signal classification in complex environments, enhances the overall reception clarity of hearing aids, and meets the personalized needs of users.
Smart Images
Figure CN121662072A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hearing aid audio classification technology, specifically a hearing aid audio classification method. Background Technology
[0002] Hearing aids are assistive devices used by people with hearing impairments to compensate for the deficiencies caused by hearing loss, thereby improving their ability to communicate with others. Hearing aids consist of a microphone, amplifier, receiver, and power supply. When wearing a hearing aid, the appropriate product should be selected based on the electroacoustic characteristics of the hearing aid and its relationship with the patient's hearing characteristics and needs. After wearing the hearing aid, it is best to get used to various background sounds in a quiet environment and try to distinguish each sound.
[0003] Traditional hearing aids can only process simple audio signals. Their ability to recognize and process audio signals in complex real-world environments is limited. Therefore, improving the audio signal processing capabilities of hearing aids to better adapt to various environments has become an urgent problem to be solved, thus facilitating their clamping and fixation. To this end, it is necessary to propose a hearing aid audio classification method. Summary of the Invention
[0004] To address the problems in the prior art, this invention provides a method for classifying audio data in hearing aids.
[0005] The technical solution adopted by the present invention to solve its technical problem is: a hearing aid audio classification method, including a classification system, wherein the classification system includes a data acquisition and preprocessing module, a neural network model construction module, a model training module, a real-time audio input module, a scene recognition stage module, a personalized adjustment module, and a feedback optimization module; The data acquisition and preprocessing module includes data preprocessing, feature extraction, and feature fusion. The data acquisition and preprocessing module collects a large amount of audio data from different scenarios and performs preprocessing. The data preprocessing involves preprocessing the collected audio data, including noise reduction, segmentation, and feature extraction. The feature extraction adopts the traditional audio feature extraction method, which extracts features from the preprocessed audio data through Mel frequency cepstral coefficients and spectral features; The feature fusion method combines the extracted traditional audio features with features extracted by deep learning methods to improve the performance of the classifier.
[0006] The neural network model building module sets up a deep neural network suitable for audio classification, performs operations by combining convolutional neural networks and recurrent neural networks, collects data as coordinates, and then builds a model using the extracted features. At the same time, an attention mechanism is added to the model to increase attention to key audio features.
[0007] Therefore, the training model module uses preprocessed audio data to train the model, and runs it with an appropriate loss function and optimization algorithm to allow the model to learn the audio features of different scenarios; Simultaneously, the model is trained iteratively multiple times, training deep learning models such as convolutional neural networks or recurrent neural networks until the module boxes achieve good performance. The real-time audio input module applies a trained neural network model to the hearing aid device to classify the real-time received audio signals. Hearing aids receive audio signals in real time and convert them into digital signals.
[0008] The scene recognition module inputs real-time digital signals into a trained model, and the model outputs a predicted category for the current audio scene.
[0009] The personalized adjustment module fine-tunes the audio classification results by combining the user's historical hearing aid usage and preferences; The weights of certain scenarios can be adjusted based on the user's specific needs.
[0010] The feedback optimization module continuously collects user feedback and makes modifications based on the feedback information for certain long-term categories. The model was further optimized and improved based on the feedback.
[0011] Deep neural network-based audio classification algorithms, such as convolutional neural networks and recurrent neural networks, can automatically learn the complex features of audio data, thereby improving classification performance. Attention-based audio classification algorithms: Attention mechanisms can help models focus on key information in audio signals, thereby improving classification accuracy; Audio classification algorithm based on transfer learning: Transfer learning can improve classification performance on small datasets by leveraging the learning results of pre-trained models on large datasets.
[0012] To evaluate and compare the performance of different audio classification methods, the following metrics can be used: Accuracy: The proportion of correctly classified samples out of the total number of samples; Recall: The proportion of correctly classified positive samples out of the actual number of positive samples; F1 score: the harmonic mean of precision and recall; Confusion matrix: Used to visualize the classification results of different classifiers for each class. Accurate audio classification is achieved using the above indicators.
[0013] Audio features are representative numerical values extracted from audio signals and used to characterize the properties of the audio signals. The main functions of audio features are as follows: Reduce data dimensionality: Audio signals are usually high-dimensional. Through feature extraction, high-dimensional data can be transformed into low-dimensional data, which is convenient for subsequent processing. Extracting useful information: Feature extraction can extract information from audio signals that is useful for classification tasks, thereby improving classification performance; Simplifying the model: Feature extraction can simplify the model structure, reduce model complexity, and improve model training efficiency.
[0014] Audio feature extraction methods: Frequency domain-based methods: Fast Fourier Transform (FFT), which can convert audio signals from the time domain to the frequency domain, thus facilitating the acquisition of the signal's spectral information; Time-domain based methods, such as short-time energy and short-time average zero-crossing rate, can directly extract features from the time-domain waveform of audio signals; frequency-time domain based methods, such as Mel-frequency cepstral coefficients, combine frequency-domain and time-domain methods to simultaneously obtain the spectral and temporal information of audio signals. This allows for accurate feature extraction using different audio feature extraction methods, facilitating subsequent audio classification. The audio feature extraction method for hearing aids needs to take into account the special characteristics of hearing aids, such as the size of the ear canal and the degree of hearing loss. Ear canal audio feature extraction: By extracting features from the ear canal audio signal, the auditory perception process of the human ear can be better simulated; Adaptive audio feature extraction: Adjust audio features according to the user's degree of hearing loss to meet the user's personalized needs; Multi-dimensional audio feature extraction: Combining audio features from multiple dimensions can more comprehensively characterize the characteristics of audio signals and improve the performance of hearing aids; By segmenting, normalizing, and enhancing audio signals, data quality is improved. Appropriate feature extraction methods, such as Mel-frequency cepstral coefficients and spectral features, are selected to fully characterize the features of the audio data. Cross-validation and other methods are used to train the model, and the classification performance is optimized by adjusting the model parameters. Evaluation metrics such as accuracy, recall, and F1 score are used to assess the model to understand its classification effect, improve the overall audio classification efficiency of hearing aids, and facilitate its use.
[0015] To ensure the reliability and generalization ability of the experimental results, a representative dataset needs to be constructed. The process of constructing and selecting the dataset includes: Data acquisition: Collecting data from different audio sources, including speech, music, noise, etc. Data preprocessing: Performing preprocessing operations such as segmentation, normalization, and enhancement on the collected data to improve data quality; Data labeling: Classifying and labeling the preprocessed audio data to provide accurate labels for model training. Dataset partitioning: The labeled dataset is divided into training, validation, and test sets to evaluate the model's generalization ability; Data filtering: Through data analysis, remove low-quality data or data that is irrelevant to the experimental task to ensure the purity and diversity of the dataset.
[0016] The method for using a hearing aid audio classification method includes the following steps: S1: Collect audio data from various scenarios to ensure data diversity and representativeness. Denoise the collected audio data to reduce noise interference in subsequent analysis. Segment the audio data for better feature extraction. Use traditional audio feature extraction methods, such as Mel frequency cepstral coefficients and spectral features, to extract features from the preprocessed audio data. Fuse the extracted traditional audio features with features extracted by deep learning methods to improve classifier performance. S2: Based on the needs of audio classification, set up a suitable deep neural network architecture, such as a combination of convolutional neural networks and recurrent neural networks. Using the collected data as coordinates and combining the extracted features, build the model structure. Add an attention mechanism to the model so that it can pay more attention to key audio features. The model is trained using preprocessed audio data. A suitable loss function and optimization algorithm are selected. Through continuous iteration, the model learns the audio features of different scenarios. The model is trained multiple times until it achieves good performance, such as high accuracy. The trained neural network model is then applied to hearing aid devices. S3: The hearing aid receives audio signals in real time, converts them into digital signals, inputs the digital signals into the trained model, and the model outputs the predicted category of the current audio scene. S4: Based on the user's historical hearing aid usage and personal preferences, fine-tune the audio classification results and adjust the weight of certain scenarios according to the user's specific needs to better meet the user's personalized needs. We continuously collect user feedback to understand the model's performance in real-world applications. For scenarios with consistently accurate classification, we make appropriate adjustments based on the feedback and further optimize and improve the model to enhance its performance and adaptability.
[0017] The beneficial effects of this invention are: The hearing aid audio classification method described in this invention combines deep learning technology with traditional audio feature extraction methods. By training a neural network model, it achieves automatic classification of audio signals received by the hearing aid. It has high accuracy and stability in hearing aid audio signal classification, thereby improving the stability of audio classification during hearing aid use and increasing the overall reception clarity of the hearing aid. Attached Figure Description
[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0019] Figure 1 This is a schematic diagram of the overall process structure of a hearing aid audio classification method provided by the present invention. Detailed Implementation
[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0021] like Figure 1 As shown, the hearing aid audio classification method of the present invention includes a classification system, which includes a data acquisition and preprocessing module, a neural network model construction module, a model training module, a real-time audio input module, a scene recognition stage module, a personalized adjustment module, and a feedback optimization module. The data acquisition and preprocessing module includes data preprocessing, feature extraction, and feature fusion. The data acquisition and preprocessing module collects a large amount of audio data from different scenarios and performs preprocessing. The data preprocessing involves preprocessing the collected audio data, including noise reduction, segmentation, and feature extraction. The feature extraction adopts the traditional audio feature extraction method, which extracts features from the preprocessed audio data through Mel frequency cepstral coefficients and spectral features; The feature fusion method combines the extracted traditional audio features with features extracted by deep learning methods to improve the performance of the classifier.
[0022] The neural network model building module sets up a deep neural network suitable for audio classification, performs operations by combining convolutional neural networks and recurrent neural networks, collects data as coordinates, and then builds a model using the extracted features. At the same time, an attention mechanism is added to the model to increase attention to key audio features.
[0023] Therefore, the training model module uses preprocessed audio data to train the model, and runs it with an appropriate loss function and optimization algorithm to allow the model to learn the audio features of different scenarios; Simultaneously, the model is trained iteratively multiple times, training deep learning models such as convolutional neural networks or recurrent neural networks until the module boxes achieve good performance.
[0024] The real-time audio input module applies a trained neural network model to the hearing aid device to classify the real-time received audio signals. Hearing aids receive audio signals in real time and convert them into digital signals.
[0025] The scene recognition module inputs real-time digital signals into a trained model, and the model outputs a predicted category for the current audio scene.
[0026] The personalized adjustment module fine-tunes the audio classification results by combining the user's historical hearing aid usage and preferences; The weights of certain scenarios can be adjusted based on the user's specific needs.
[0027] The feedback optimization module continuously collects user feedback and makes modifications based on the feedback information for certain long-term categories. The model was further optimized and improved based on the feedback.
[0028] Deep neural network-based audio classification algorithms, such as convolutional neural networks and recurrent neural networks, can automatically learn the complex features of audio data, thereby improving classification performance. Attention-based audio classification algorithms: Attention mechanisms can help models focus on key information in audio signals, thereby improving classification accuracy; Audio classification algorithm based on transfer learning: Transfer learning can improve classification performance on small datasets by leveraging the learning results of pre-trained models on large datasets.
[0029] To evaluate and compare the performance of different audio classification methods, the following metrics can be used: Accuracy: The proportion of correctly classified samples out of the total number of samples; Recall: The proportion of correctly classified positive samples out of the actual number of positive samples; F1 score: the harmonic mean of precision and recall; Confusion matrix: Used to visualize the classification results of different classifiers for each class. Accurate audio classification is achieved using the above indicators.
[0030] Audio features are representative numerical values extracted from audio signals and used to characterize the properties of the audio signals. The main functions of audio features are as follows: Reduce data dimensionality: Audio signals are usually high-dimensional. Through feature extraction, high-dimensional data can be transformed into low-dimensional data, which is convenient for subsequent processing. Extracting useful information: Feature extraction can extract information from audio signals that is useful for classification tasks, thereby improving classification performance; Simplifying the model: Feature extraction can simplify the model structure, reduce model complexity, and improve model training efficiency.
[0031] Audio feature extraction methods: Frequency domain-based methods: Fast Fourier Transform (FFT), which can convert audio signals from the time domain to the frequency domain, thus facilitating the acquisition of the signal's spectral information; Time-domain based methods, such as short-time energy and short-time average zero-crossing rate, can directly extract features from the time-domain waveform of audio signals; frequency-time domain based methods, such as Mel-frequency cepstral coefficients, combine frequency-domain and time-domain methods to simultaneously obtain the spectral and temporal information of audio signals. This allows for accurate feature extraction using different audio feature extraction methods, facilitating subsequent audio classification. The audio feature extraction method for hearing aids needs to take into account the special characteristics of hearing aids, such as the size of the ear canal and the degree of hearing loss. Ear canal audio feature extraction: By extracting features from the ear canal audio signal, the auditory perception process of the human ear can be better simulated; Adaptive audio feature extraction: Adjust audio features according to the user's degree of hearing loss to meet the user's personalized needs; Multi-dimensional audio feature extraction: Combining audio features from multiple dimensions can more comprehensively characterize the characteristics of audio signals and improve the performance of hearing aids; By segmenting, normalizing, and enhancing audio signals, data quality is improved. Appropriate feature extraction methods, such as Mel-frequency cepstral coefficients and spectral features, are selected to fully characterize the features of the audio data. Cross-validation and other methods are used to train the model, and the classification performance is optimized by adjusting the model parameters. Evaluation metrics such as accuracy, recall, and F1 score are used to assess the model to understand its classification effect, improve the overall audio classification efficiency of hearing aids, and facilitate its use.
[0032] To ensure the reliability and generalization ability of the experimental results, a representative dataset needs to be constructed. The process of constructing and selecting the dataset includes: Data acquisition: Collecting data from different audio sources, including speech, music, noise, etc. Data preprocessing: Performing preprocessing operations such as segmentation, normalization, and enhancement on the collected data to improve data quality; Data labeling: Classifying and labeling the preprocessed audio data to provide accurate labels for model training. Dataset partitioning: The labeled dataset is divided into training, validation, and test sets to evaluate the model's generalization ability; Data filtering: Through data analysis, remove low-quality data or data that is irrelevant to the experimental task to ensure the purity and diversity of the dataset.
[0033] Specific applications include: extensively collecting audio data from various scenarios to ensure data diversity and representativeness; denoising the collected audio data to reduce noise interference in subsequent analysis; segmenting the audio data for better feature extraction; employing traditional audio feature extraction methods, such as Mel-frequency cepstral coefficients and spectral features, to extract features from the preprocessed audio data; fusing the extracted traditional audio features with features extracted by deep learning methods to improve classifier performance; setting up appropriate deep neural network architectures based on audio classification needs, such as a combination of convolutional neural networks and recurrent neural networks; building the model structure using the collected data as coordinates and combining it with the extracted features; incorporating an attention mechanism into the model to enable it to focus more on key audio features; training the model using the preprocessed audio data and selecting an appropriate loss function. The algorithm is optimized through continuous iteration, allowing the model to learn the audio features of different scenarios. Multiple iterations of training are performed until the model achieves good performance, such as high accuracy. The trained neural network model is then applied to hearing aid devices. The hearing aid receives audio signals in real time, converts them into digital signals, and inputs these digital signals into the trained model. The model outputs a predicted category for the current audio scenario. The audio classification results are fine-tuned based on the user's historical hearing aid usage and personal preferences. Weights for certain scenarios are adjusted according to the user's specific needs to better meet individual requirements. User feedback is continuously collected to understand the model's performance in practical applications. For scenarios with consistently high classification accuracy, appropriate adjustments are made based on feedback. The model is further optimized and improved based on feedback to enhance its performance and adaptability.
[0034] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for using a hearing aid audio classification method, characterized in that: Includes the following steps: first step: When it is necessary to classify hearing aid audio, audio data from different scenarios is collected extensively to ensure the diversity and representativeness of the data. The collected audio data is denoised to reduce the interference of noise on subsequent analysis. The audio data is segmented to facilitate better feature extraction. Traditional audio feature extraction methods, such as Mel frequency cepstral coefficients and spectral features, are used to extract features from the preprocessed audio data. The extracted traditional audio features are then fused with features extracted by deep learning methods to improve the classifier performance. Step 2: Based on the needs of audio classification, set up a suitable deep neural network architecture, such as a combination of convolutional neural networks and recurrent neural networks. Using the collected data as coordinates and combining the extracted features, build the model structure. Add an attention mechanism to the model so that it can pay more attention to key audio features. The model is trained using preprocessed audio data. A suitable loss function and optimization algorithm are selected. Through continuous iteration, the model learns the audio features of different scenarios. The model is trained multiple times until it achieves good performance, such as high accuracy. The trained neural network model is then applied to hearing aid devices. Step 3: The hearing aid receives audio signals in real time, converts them into digital signals, inputs the digital signals into the trained model, and the model outputs the predicted category of the current audio scene. Step 4: Based on the user's historical hearing aid usage and personal preferences, fine-tune the audio classification results, and adjust the weight of certain scenarios according to the user's specific needs to better meet the user's personalized needs. We continuously collect user feedback to understand the model's performance in real-world applications. For scenarios with consistently accurate classification, we make appropriate adjustments based on the feedback and further optimize and improve the model to enhance its performance and adaptability.
2. A method for classifying audio data of hearing aids, characterized in that, The system includes a classification system, which comprises a data acquisition and preprocessing module, a neural network model construction module, a model training module, a real-time audio input module, a scene recognition module, a personalized adjustment module, and a feedback optimization module. The data acquisition and preprocessing module includes data preprocessing, feature extraction, and feature fusion. The data acquisition and preprocessing module collects a large amount of audio data from different scenarios and performs preprocessing. The data preprocessing involves preprocessing the collected audio data, including noise reduction, segmentation, and feature extraction. The feature extraction adopts the traditional audio feature extraction method, which extracts features from the preprocessed audio data through Mel frequency cepstral coefficients and spectral features; The feature fusion method combines the extracted traditional audio features with features extracted by deep learning methods to improve the performance of the classifier.
3. The hearing aid audio classification method according to claim 1, characterized in that: The neural network model building module sets up a deep neural network suitable for audio classification, performs operations by combining convolutional neural networks and recurrent neural networks, collects data as coordinates, and then builds a model using the extracted features. At the same time, an attention mechanism is added to the model to increase attention to key audio features.
4. The hearing aid audio classification method according to claim 1, characterized in that: Therefore, the training model module uses preprocessed audio data to train the model, and runs it with an appropriate loss function and optimization algorithm to allow the model to learn the audio features of different scenarios; Simultaneously, the model is trained iteratively multiple times, training deep learning models such as convolutional neural networks or recurrent neural networks until the module boxes achieve good performance.
5. The hearing aid audio classification method according to claim 1, characterized in that: The real-time audio input module applies a trained neural network model to the hearing aid device to classify the real-time received audio signals. Hearing aids receive audio signals in real time and convert them into digital signals.
6. The hearing aid audio classification method according to claim 1, characterized in that: The scene recognition module inputs real-time digital signals into a trained model, and the model outputs a predicted category for the current audio scene.
7. The hearing aid audio classification method according to claim 1, characterized in that: The personalized adjustment module fine-tunes the audio classification results by combining the user's historical hearing aid usage and preferences; The weights of certain scenarios can be adjusted based on the user's specific needs.
8. The hearing aid audio classification method according to claim 1, characterized in that: The feedback optimization module continuously collects user feedback and makes modifications based on the feedback information for certain long-term categories. The model was further optimized and improved based on the feedback.