Human ear perception frequency feature code retrieval method

Abstract

The invention provides a human ear perception frequency feature code retrieval method, which belongs to the technical field of voice processing, and comprises the following steps: audio signal preprocessing: digitalizing original audio by adopting a proper sampling rate and bit depth, and applying a noise filtering technology, such as wavelet denoising or spectral subtraction, to obtain an audio signal; according to the steps of background noise interference reduction, feature selection and dimension reduction, most representative features are screened, and PCA, LDA and other dimension reduction technologies are applied, so that the calculation complexity is reduced, the processing speed is improved, dimension disasters are avoided, the stability and accuracy of the algorithm are ensured, and the method is particularly important for processing large-scale audio data sets.

Description

Technical Field

[0001] The invention belongs to the technical field of speech processing, and in particular relates to a method for retrieving a frequency characteristic code perceived by human ears. Background Art

[0002] In the field of audio signal processing and retrieval, with the rapid development of the Internet and multimedia technology, the amount and complexity of audio data are growing day by day. How to effectively manage and retrieve these massive audio resources has become an urgent problem to be solved. Traditional audio retrieval methods mainly rely on text-based metadata (such as title, artist, album information, etc.) or simple audio signal features. These methods often cannot fully meet users' precise search needs for audio content details, especially when dealing with complex audio types such as speech and music, and lack a mechanism for in-depth understanding and perception of the audio content itself.

[0003] The human ear's perception of audio signals is highly nonlinear and complex, and it can keenly capture subtle changes in different frequencies, rhythms, intensities, etc., but existing technologies have limitations in simulating the perceptual characteristics of the human ear. For example, standard audio feature extraction technology may not fully consider the differences in the human ear's response to different frequencies, resulting in the neglect of many audio features that are significant to the human ear but difficult to capture by traditional methods during the retrieval process. In addition, factors such as noise interference, the wide dynamic range of audio signals, and high real-time requirements are also important challenges faced by existing technologies.

[0004] In view of the above problems, existing audio retrieval systems often fail to provide efficient, accurate retrieval results that are consistent with human ear perception habits, especially in application scenarios such as classification, recognition, and personalized recommendation of large amounts of audio data. Therefore, there is an urgent need for a new method for retrieval of human ear perception frequency feature codes, which should be able to effectively simulate the human auditory system and extract features that are highly consistent with human ear perception from complex audio signals, while also having the ability to quickly process and retrieve audio data to meet the needs of modern audio information processing and applications.

[0005] Therefore, based on the above technical problems, it is necessary to design a method for retrieving frequency feature codes perceived by the human ear. Summary of the invention

[0006] The purpose of the present invention is to provide a method for retrieving frequency feature codes perceived by the human ear, aiming to solve the problem that existing audio retrieval systems in the prior art often fail to provide efficient, accurate retrieval results that conform to the human ear's perception habits, especially in application scenarios such as classification, recognition and personalized recommendation of large amounts of audio data.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for retrieving a human ear perception frequency characteristic code, characterized in that it comprises the following steps:

[0009] A1. Audio signal preprocessing: Digitize the original audio using appropriate sampling rate and bit depth, and apply noise filtering techniques such as wavelet denoising or spectral subtraction to reduce background noise interference;

[0010] B2. Perception-oriented feature extraction:

[0011] Mel frequency cepstral coefficient: simulates the nonlinear characteristics of human ear's response to different frequencies and extracts spectrum envelope characteristics;

[0012] Peak frequency detection: Identify prominent frequency components in the audio, corresponding to the frequencies that the human ear is sensitive to;

[0013] Rhythm and intensity analysis: Analyze the rhythm pattern and intensity changes of the audio to supplement the time domain features;

[0014] C3. Feature selection and dimensionality reduction: Screen features based on their importance and retain the most discriminative features; apply dimensionality reduction techniques (such as PCA and LDA) to reduce the dimension of the feature space and increase processing speed.

[0015] As a preferred solution of the present invention, D4, dynamic perception adaptive processing:

[0016] Adaptive adjustment of dynamic range ensures balanced processing of audio with different loudness;

[0017] Application of frequency masking models to exclude audio components that are not perceptually significant

[0018] E5. Deep Learning Enhancement:

[0019] Introducing the time-frequency attention mechanism to intelligently allocate processing resources and focus on key audio segments;

[0020] Use deep neural networks for sentiment analysis and style recognition to enrich feature dimensions;

[0021] F6 creates feature database and index:

[0022] Efficiently store the processed feature codes and use inverted indexes or hash tables to accelerate retrieval;

[0023] G7, real-time retrieval and matching algorithm:

[0024] Develop real-time feature extraction algorithms to support instant retrieval of audio streams;

[0025] Optimize network transmission to ensure fast transmission and low-latency retrieval of high-quality audio.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. In this scheme, through reasonable audio signal preprocessing (A1), including adaptive sampling rate and bit depth setting and the application of noise filtering technology, high-quality digitization of audio data is ensured, laying a solid foundation for subsequent processing. Perception-oriented feature extraction (B2), especially the use of Mel-frequency cepstral coefficients (MFCC), as well as peak frequency detection and rhythm and intensity analysis, closely simulates the nonlinear perception characteristics of the human ear, extracts key features consistent with human ear perception, and improves the depth of understanding of audio content. The feature selection and dimensionality reduction step (C3) not only reduces the computational complexity and improves the processing speed by screening the most representative features and applying dimensionality reduction techniques such as PCA and LDA, but also avoids the dimensionality disaster and ensures the stability and accuracy of the algorithm, which is particularly important for processing large-scale audio data sets.

[0028] 2. In this solution, dynamic perception adaptive processing (D4) and deep learning enhancement (E5) strategies, such as dynamic range adjustment, application of frequency masking model, introduction of time-frequency attention mechanism, as well as sentiment analysis and style recognition, not only enable the system to adapt to the characteristics of different audios, but also deepen the analysis of audio content through deep learning technology, enrich the feature dimensions, enhance personalized retrieval capabilities, and improve user experience.

[0029] 3. In this solution, the establishment of feature database and index (F6) and the design of real-time retrieval and matching algorithm (G7) ensure instant retrieval even for streaming audio through efficient storage structure and fast retrieval technology, such as inverted index and hash table, as well as real-time feature extraction algorithm and network transmission optimization, thus meeting the requirements of modern audio applications for low latency and fast response. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0031] Figure 1 This is a flow chart of the human ear perception frequency feature code retrieval method of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Example 1

[0034] See also Figure 1 , the present invention provides the following technical solutions:

[0035] A method for retrieving a frequency characteristic code perceived by human ears comprises the following steps:

[0036] A1. Audio signal preprocessing: Digitize the original audio using appropriate sampling rate and bit depth, and apply noise filtering techniques such as wavelet denoising or spectral subtraction to reduce background noise interference;

[0037] B2. Perception-oriented feature extraction:

[0038] Mel frequency cepstral coefficient: simulates the nonlinear characteristics of human ear's response to different frequencies and extracts spectrum envelope characteristics;

[0039] Peak frequency detection: Identify prominent frequency components in the audio, corresponding to the frequencies that the human ear is sensitive to;

[0040] Rhythm and intensity analysis: Analyze the rhythm pattern and intensity changes of the audio to supplement the time domain features;

[0041] C3. Feature selection and dimensionality reduction: Screen features based on their importance and retain the most discriminative features; apply dimensionality reduction techniques (such as PCA and LDA) to reduce the dimension of the feature space and increase processing speed.

[0042] In a specific embodiment of the present invention, A1 audio signal preprocessing, this step ensures that the audio data can be accurately and efficiently analyzed and processed by the system. By selecting a suitable sampling rate and bit depth to digitize the original audio, it can ensure that the information is converted into a digital signal without distortion. Noise filtering technology, such as wavelet denoising or spectral subtraction, is applied to effectively filter out background noise in the recording environment, so that subsequent processing can focus more on the content of the audio signal itself and improve signal purity. B2 perception-oriented feature extraction, Mel-frequency cepstral coefficient (MFCC): using this technology, the nonlinear response of the human ear to sounds of different frequencies is simulated Characteristics, extract the spectral envelope characteristics of the audio signal. MFCC is a commonly used feature representation method in speech and music analysis. It can well reflect the auditory perception characteristics of the human ear and is very effective in distinguishing different timbres, speakers, etc. Peak frequency detection: By identifying the frequency point (peak frequency) with the densest energy distribution in the audio signal, the audio component that the human ear is most sensitive to is captured. This is crucial for identifying specific musical instruments, voice fundamental frequencies, etc. Rhythm and intensity analysis: Analyze the rhythm pattern (such as beat, speed) and intensity (volume) changes of the audio to provide time domain information for the feature code, which helps to distinguish different The music of different styles and the emotional expression of language, etc., supplement another dimension of information besides the frequency domain features. C3 feature selection and dimensionality reduction. After completing the feature extraction, facing the high-dimensional feature space, the present invention adopts feature selection and dimensionality reduction technology to optimize the processing flow: Feature selection: Screen according to the contribution of the feature in distinguishing different audio contents, retain the most representative and distinguishing features, and remove redundant information. Dimensionality reduction technology application: Use principal component analysis (PCA), linear discriminant analysis (LDA) and other technologies to further reduce the feature dimension, while trying to maintain the diversity of data. This not only greatly speeds up the subsequent processing and retrieval speed, but also avoids the "dimensional disaster" and improves the stability and accuracy of the algorithm. Through the above-mentioned scheme, efficient and accurate processing of audio signals is achieved, and data quality is ensured from the preprocessing stage to the perception-oriented feature extraction, which is closely in line with the human ear's auditory perception. Finally, through intelligent feature selection and dimensionality reduction strategies, not only the processing speed is improved, but also the accuracy and robustness of audio retrieval are enhanced. This method is particularly suitable for the management and retrieval system of large-scale audio databases, such as music recommendation, speech recognition, environmental sound analysis and other fields, and can effectively improve user experience and service quality.

[0043] For details, please refer to Figure 1 , D4, dynamic perception adaptive processing:

[0044] Adaptive adjustment of dynamic range ensures balanced processing of audio with different loudness;

[0045] Application of frequency masking models to exclude audio components that are not perceptually significant

[0046] E5. Deep Learning Enhancement:

[0047] Introducing the time-frequency attention mechanism to intelligently allocate processing resources and focus on key audio segments;

[0048] Use deep neural networks for sentiment analysis and style recognition to enrich feature dimensions;

[0049] F6 creates feature database and index:

[0050] Efficiently store the processed feature codes and use inverted indexes or hash tables to accelerate retrieval;

[0051] G7, real-time retrieval and matching algorithm:

[0052] Develop real-time feature extraction algorithms to support instant retrieval of audio streams;

[0053] Optimize network transmission to ensure fast transmission and low-latency retrieval of high-quality audio.

[0054] In this embodiment: Adaptive adjustment of dynamic range: Automatically adjust processing parameters to adapt to audio inputs of different loudness, ensuring that both gentle whispers and passionate music can be processed in a balanced and appropriate manner, improving the consistency of processing in diverse audio content, Application of frequency masking model: Simulate the auditory masking effect of the human ear, that is, a strong signal will mask the adjacent weak signal, by excluding audio components that are not significant or masked in human ear perception, and concentrate on processing information that has a significant impact on auditory perception, thereby improving the effectiveness and accuracy of processing, E5 Deep learning enhancement, time-frequency attention mechanism: With the help of deep learning technology, an attention mechanism is introduced to intelligently allocate processing resources. This mechanism can dynamically focus on the most critical or information-rich parts of the audio on the time axis and frequency axis, optimize the utilization of computing resources, and enhance the ability to capture complex audio features, Emotional analysis and style recognition: Use deep neural networks to conduct a deeper analysis of audio, not only limited to basic physical features, but also to identify emotional colors and style features in audio, adding a new dimension to audio retrieval, enabling the system to provide more personalized search results based on the user's emotional preferences or style tendencies, F6 Establish Feature database and index, build an efficient storage structure, organize the feature codes obtained through the above processing, and use efficient indexing technology (such as inverted index or hash table) to accelerate the retrieval process. This design greatly improves the efficiency of finding specified audio features in large-scale databases. G7 real-time retrieval and matching algorithm, real-time feature extraction algorithm: The developed algorithm can extract features instantly during audio streaming without waiting for the entire audio file to be loaded, meeting the needs of real-time interaction and monitoring scenarios, and optimizing network transmission: By optimizing data transmission protocols and compression technologies, the rapid transmission of high-quality audio data is ensured, while reducing network latency, improving the user experience when performing audio retrieval and playback, especially in live broadcasts and online music services that require low latency. The technical solution builds a comprehensive, intelligent and efficient audio processing and retrieval system through dynamic perception adaptive processing, deep learning enhancement, efficient database management and real-time retrieval technology. It not only improves the accuracy and robustness of audio processing, but also greatly enhances the user experience, especially in processing complex and changeable audio application scenarios with high real-time requirements, showing significant technical advantages and application value.

[0055] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for retrieving a frequency characteristic code perceived by human ears, characterized in that: It includes the following steps: A1. Audio signal preprocessing: Digitize the original audio using appropriate sampling rate and bit depth, and apply noise filtering techniques such as wavelet denoising or spectral subtraction to reduce background noise interference; B2. Perception-oriented feature extraction: Mel frequency cepstral coefficient: simulates the nonlinear characteristics of human ear's response to different frequencies and extracts spectrum envelope characteristics; Peak frequency detection: Identify prominent frequency components in the audio, corresponding to the frequencies that the human ear is sensitive to; Rhythm and intensity analysis: Analyze the rhythm pattern and intensity changes of the audio to supplement the time domain features; C3. Feature selection and dimensionality reduction: Screen features based on their importance and retain the most discriminative features; apply dimensionality reduction techniques (such as PCA and LDA) to reduce the dimension of the feature space and increase processing speed.

2. The method for retrieving a human ear perception frequency characteristic code according to claim 1, further comprising: D4, Dynamic Perception Adaptive Processing: Adaptive adjustment of dynamic range ensures balanced processing of audio with different loudness; Application of frequency masking models to exclude audio components that are not perceptually significant E5. Deep Learning Enhancement: Introducing the time-frequency attention mechanism to intelligently allocate processing resources and focus on key audio segments; Use deep neural networks for sentiment analysis and style recognition to enrich feature dimensions; F6 creates feature database and index: Efficiently store the processed feature codes and use inverted indexes or hash tables to accelerate retrieval; G7, real-time retrieval and matching algorithm: Develop real-time feature extraction algorithms to support instant retrieval of audio streams; Optimize network transmission to ensure fast transmission and low-latency retrieval of high-quality audio.

Images