Audio classification method and device, electronic equipment and storage medium

Abstract

The invention relates to an audio classification method. The method comprises the following steps: extracting features of a plurality of audio signals to be classified; based on the feature of each to-be-classified audio signal in the plurality of to-be-classified audio signals, performing initial classification on the plurality of to-be-classified audio signals to obtain a plurality of initial audio signal sets; each initial audio signal set comprises one or more to-be-classified audio signals; performing clustering processing on one or more to-be-classified audio signals in each initial audio signal set to obtain a plurality of target audio signal sets; wherein each initial audio signal set corresponds to one or more target audio signal sets. The embodiment of the invention further provides an audio classification device, electronic equipment and a storage medium.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile audio processing, and in particular to an audio classification method, device, electronic equipment and storage medium. Background Art

[0002] Cars are currently the main means of transportation, and users spend 1 / 6 to 1 / 3 of their time in cars every day. In-car music provides an important entertainment program for users. As more and more people own cars, users have higher requirements for car music. However, current car audio systems can only simply classify audio into several basic types, and can only roughly classify music. Each type of music style contains a relatively large range of music styles, and cannot identify the specific style characteristics of car audio. As a result, when users play music, there are relatively few audio style options provided to users, and they cannot find the audio style they want, thereby reducing the user experience. Summary of the invention

[0003] One of the purposes of the present invention is to provide an audio classification method to solve the problem that when users play music, there are relatively few audio style options provided to users, and they cannot find the desired audio style, thereby reducing the user experience; the second purpose is to provide an audio classification device; the third purpose is to provide an electronic device and a storage medium.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] Extract features of multiple audio signals to be classified; based on the features of each of the multiple audio signals to be classified, initially classify the multiple audio signals to be classified to obtain multiple initial audio signal sets; each initial audio signal set includes one or more audio signals to be classified; cluster the one or more audio signals to be classified in each initial audio signal set to obtain multiple target audio signal sets; wherein each initial audio signal set corresponds to one or more target audio signal sets.

[0006] According to the above technical means, by roughly classifying multiple audio signals to be classified according to the characteristics of each audio signal to be classified, multiple initial audio signal sets are obtained, and then each initial audio signal set is clustered to obtain multiple target audio signal sets, wherein each initial audio signal set includes one or more audio signals to be classified; each initial audio signal set corresponds to one or more target audio signal sets. In this way, by first roughly classifying multiple audio signals to be classified, and then finely classifying each initial audio signal set obtained after the rough classification to obtain a target audio signal set, it is possible to finely classify the characteristics of the audio signals to be classified, and obtain a variety of rich audio style categories, thereby solving the problem that users cannot find the desired music style when playing music.

[0007] Further, the features include: time-frequency features and / or attribute features; based on the time-frequency features and / or attribute features of each of the multiple audio signals to be classified, the multiple audio signals to be classified are initially classified to obtain the multiple initial audio signal sets.

[0008] According to the above technical means, multiple audio signals to be classified are initially classified according to the time-frequency characteristics and / or attribute characteristics of each audio signal to be classified, and multiple initial audio signal sets are obtained. In this way, the time-frequency characteristics can provide a deeper understanding of the information of the audio signals to be classified, and the attribute characteristics can intuitively understand the surface information of the audio signals to be classified. The time-frequency characteristics and / or attribute characteristics are used to comprehensively understand the information of the audio signals to be classified, which is conducive to better classification of the audio signals to be classified and obtaining more abundant audio categories.

[0009] Further, the time-frequency features include: time domain features and / or frequency domain features; based on the time domain features and / or frequency domain features of each audio signal to be classified, the multiple audio signals to be classified are initially classified for the first time to obtain multiple second audio signal sets; each second audio signal set includes one or more audio signals to be classified; based on the attribute features of each audio signal to be classified, one or more audio signals to be classified in each second audio signal set are initially classified for the second time to obtain the multiple initial audio signal sets; wherein each initial audio signal set corresponds to one or more second audio signal sets; wherein the attribute features include one or more of the following: pitch, duration, timbre, speed and intensity.

[0010] According to the above technical means, the time domain features and / or frequency domain features of each audio signal to be classified are first used to perform an initial classification of multiple audio signals to be classified, and then the second audio signal set obtained after the first initial classification is classified for a second time according to the attribute features of each audio signal to be classified, such as pitch, duration, timbre, speed and strength, to obtain multiple initial audio signal sets. In this way, through two classifications, one or more audio signals to be classified in the obtained initial audio signal sets can be made more identical or similar, which is beneficial to the subsequent classification of audio signals and improves the accuracy of the classification of audio signals to be classified.

[0011] Further, K cluster centers in each initial audio signal set are determined; K is an integer greater than or equal to 1; based on the K cluster centers, K cluster sets corresponding to the K cluster centers are determined; when the similarity between the K cluster sets and the K-1 cluster sets corresponding to the K-1 cluster centers satisfies the first condition, the K cluster sets are determined as K target audio signal sets.

[0012] According to the above-mentioned technical means, one or more audio signals to be classified in each initial audio signal set are clustered, and in the process of clustering, by continuously changing the number of cluster centers, it is determined that when the similarity between K cluster sets determined by K cluster centers and K-1 cluster sets determined by K-1 cluster centers satisfies the first condition, the K cluster sets are determined as K target audio signal sets. In this way, when the number of cluster centers is changed, one or more audio signals to be classified in each cluster set can be continuously updated, thereby playing a screening role. When the similarity satisfies the first condition, it indicates that the cluster sets converge to achieve the most accurate classification effect.

[0013] Further, based on the K cluster sets, a first intra-class distance is determined; the first intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K cluster sets to the corresponding cluster center; based on the K-1 cluster sets, a second intra-class distance is determined; the second intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K-1 cluster sets to the corresponding cluster center; based on the first intra-class distance and the second intra-class distance, it is determined whether the similarity meets the first condition.

[0014] According to the above technical means, whether the similarity satisfies the first condition is determined based on the first intra-class distance obtained by adding the square of the distance from each audio signal to be classified in the K cluster sets to the corresponding cluster center, and the second intra-class distance obtained by adding the square of the distance from each audio signal to be classified in the K-1 cluster sets to the corresponding cluster center. In this way, by using the first intra-class distance and the second intra-class distance, it is possible to accurately judge whether the K cluster sets determined by the K cluster centers are in a convergence state and whether the most accurate classification effect is achieved, thereby improving the accuracy of clustering.

[0015] Further, based on the first intra-class distance, a first ratio is determined; wherein the first ratio is the ratio of the first inter-class distance to the first intra-class distance, and the first inter-class distance is obtained by adding the sum of the squares of the distances between the K cluster centers and the centers of the corresponding initial audio signal set; based on the second intra-class distance, a second ratio is determined; wherein the second ratio is the ratio of the second inter-class distance to the second intra-class distance, and the second inter-class distance is obtained by adding the sum of the squares of the distances between the K-1 cluster centers and the centers of the corresponding initial audio signal set; based on the difference between the first ratio and the second ratio, it is determined whether the similarity meets the first condition.

[0016] According to the above technical means, the difference between the ratio of the first inter-class distance and the intra-class distance, and the ratio of the second inter-class distance and the first class distance is used to determine whether the similarity meets the first condition. The accuracy of the classification of the audio signals to be classified in the cluster set is determined by utilizing both the inter-class distance and the intra-class distance, thereby improving the accuracy of clustering and further improving the accuracy of audio classification.

[0017] Further, the multiple initial audio signals to be classified are obtained; each initial audio signal to be classified is standardized to obtain each audio signal to be classified; the standardized processing includes one or more of the following: averaging multi-channel data, removing silent segment signals of audio, and normalizing signals.

[0018] According to the above technical means, after obtaining multiple initial audio signals to be classified, multi-averaging multi-channel data is performed on each initial audio signal to be classified, and the silent segment signal of the audio and one or more standardization processes in the normalization process of the signal are removed to avoid the existence of multiple signals with large differences in the initial audio signal to be classified, which affects the accuracy of subsequent classification of the audio signal.

[0019] An electronic device includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor and executes the audio classification method provided above.

[0020] A computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the audio classification method provided above.

[0021] Beneficial effects of the present invention:

[0022] By first roughly classifying the audio signals to be classified, and then finely classifying each initial audio signal set obtained after the rough classification, a target audio signal set is obtained. It is possible to perform fine classification based on the characteristics of the audio signals to be classified, and obtain a variety of rich audio style categories, further providing users with more audio style categories for users to choose from, solving the problem of users not finding the desired music style when playing music. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 An audio classification method process provided by an embodiment of the present invention Figure 1 ;

[0024] Figure 2 An audio classification method process provided by an embodiment of the present invention Figure 2 ;

[0025] Figure 3 An audio classification method process provided by an embodiment of the present invention Figure 3 ;

[0026] Figure 4 An audio classification method process provided by an embodiment of the present invention Figure 4 ;

[0027] Figure 5 A flow chart of a clustering method provided by an embodiment of the present invention;

[0028] Figure 6 An audio classification method process provided by an embodiment of the present invention Figure 5 ;

[0029] Figure 7 An audio classification method process provided by an embodiment of the present invention Figure 6 ;

[0030] Figure 8 A schematic diagram of the structure of an audio classification device provided by an embodiment of the present invention;

[0031] Fig. 9 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0033] To facilitate understanding of the technical solution of the present invention, the related technologies of the present invention are described below. The following related technologies can be arbitrarily combined with the technical solution of the present invention as optional solutions, and they all belong to the protection scope of the present invention.

[0034] In addition, in the embodiments of the present invention, "first", "second", etc. are used to distinguish similar objects, and are not necessarily used for a specific order or sequence.

[0035] As users become more and more dependent on cars for travel, the in-car audio system has become an important thing in the user's ride. Playing music through the in-car audio system can relieve the user's fatigue and boredom during the ride, greatly reducing the user's pain during long-term rides.

[0036] However, the current car audio system can only simply classify audio into several basic types, and cannot identify the specific style characteristics of car audio. It can only roughly classify music. The range of music styles contained in each type of music style is relatively large, resulting in users being unable to find the desired music style when playing music, thereby reducing the user experience. For example, the current car audio system can only classify music into songs from the 70s, 80s, and 90s, but cannot classify music from the 70s into happy, sad, emotional, etc., so when users play music, the music style is monotonous and they cannot find the desired music style. Moreover, after playing music, the user cannot recommend music of similar style to the user based on the self-learning algorithm.

[0037] Based on the related technical problems, the present invention provides an audio classification method, which roughly classifies multiple audio signals to be classified according to the characteristics of each audio signal to be classified to obtain multiple initial audio signal sets, and then clusters each initial audio signal set to obtain multiple target audio signal sets, wherein each initial audio signal set includes one or more audio signals to be classified; each initial audio signal set corresponds to one or more target audio signal sets. In this way, by first roughly classifying multiple audio signals to be classified, and then finely classifying each initial audio signal set obtained after the rough classification to obtain a target audio signal set, it is possible to finely classify the characteristics of the audio signals to be classified, obtain a variety of rich audio style categories, and further provide users with more audio style categories for users to choose from, solving the problem that users cannot find the desired music style when playing music.

[0038] The audio classification algorithm provided by the present invention can be used in the fields of vehicle-mounted, sound processing, sound interaction, sound operation, etc. It can also be applied to any electronic device that can play music, including but not limited to mobile phones, computers, tablets, wearable electronic devices, etc. The present invention does not make specific limitations on this.

[0039] To facilitate understanding of the technical solution of the present invention, the technical solution of the present invention is described in detail below through specific embodiments. The above related technologies can be combined with the technical solution of the present invention as optional solutions, and they all belong to the protection scope of the present invention. The present invention includes at least part of the following contents.

[0040] In the embodiment of the present invention, Figure 1 As shown, Figure 1 : is a flow chart of an audio classification method provided by the present invention, the audio classification method comprising:

[0041] Step S110: extracting features of a plurality of audio signals to be classified.

[0042] Step S120: Based on the feature of each of the multiple audio signals to be classified, initially classify the multiple audio signals to be classified to obtain multiple initial audio signal sets; each initial audio signal set includes one or more audio signals to be classified.

[0043] Step S130: performing clustering processing on one or more audio signals to be classified in each of the initial audio signal sets to obtain multiple target audio signal sets; wherein each of the initial audio signal sets corresponds to one or more target audio signal sets.

[0044] It should be noted that in the embodiment of the present invention, multiple audio signals to be classified are classified according to the characteristics of the audio signals to be classified. The categories here may include categories of audio styles, such as cheerful, bright, soothing, fast, quiet, etc., or they may be classified according to the atmosphere of the audio, such as warm, entertaining, leisure, etc., or they may be classified according to the age of the audio, such as 70s, 80s, 90s, etc. The embodiment of the present invention does not specifically limit the classification categories, and the audio classification method provided by the embodiment of the present invention can be used for the classification of audio.

[0045] In an embodiment of the present invention, the audio signal to be classified may be any available audio signal, including but not limited to audio signals from online music in the car, music from a CD, music imported from a USB flash drive, music in a radio channel, online music in a mobile phone, music in WeChat or telephone voice, and audio information in navigation. After obtaining the audio signal to be classified, multiple audio signals to be classified may be stored in a folder to construct an audio data set, and the embodiment of the present invention does not impose any limitation on this.

[0046] In an embodiment of the present invention, methods for obtaining features of multiple audio signals to be classified include but are not limited to convolutional neural networks, power spectrum, inverse spectrum, envelope spectrum, short-time energy, wavelet transform, short-time zero-crossing rate, etc. The embodiment of the present invention does not limit the method for obtaining the features of audio signals.

[0047] In an embodiment of the present invention, an initial audio signal set refers to a category of audio signals to be classified obtained by initial classification, each initial audio signal set represents a type of feature, and based on the feature of each audio signal to be classified, multiple audio signals to be classified are classified, and one or more audio signals to be classified with the same feature category are classified into the same initial audio signal set.

[0048] In the embodiment of the present invention, the multiple initial audio signal sets refer to multiple categories obtained by classifying multiple audio signals to be classified, and each initial audio signal set includes one or more audio signals to be classified.

[0049] In the embodiment of the present invention, the initial classification method includes but is not limited to a neural network, a support vector machine, a decision tree classification, a naive Bayes classifier, etc., and the embodiment of the present invention does not impose any specific limitation on this.

[0050] In the embodiment of the present invention, the clustering processing method can adopt K-means clustering, mean shift clustering algorithm, hierarchical clustering, density-based clustering method, maximum expectation clustering using Gaussian mixture model, etc., and the embodiment of the present invention does not make specific restrictions on this.

[0051] Exemplarily, the present invention can use the K-means clustering algorithm to cluster one or more audio signals to be classified in each initial audio signal set. The first step is to determine K cluster centers, i.e., the number of division categories, in each initial audio signal set, and randomly select K audio signals to be classified as initial cluster centers from one or more audio signals to be classified in each initial audio signal set; the second step is to calculate the distance from each audio signal to be classified in each initial audio signal set to the K initial cluster centers, assign each audio signal to be classified to the initial cluster center with the closest distance, and divide the one or more audio signals to be classified in each initial audio signal set into K initial cluster sets; the third step is to then determine the average value of multiple audio signals to be classified in each initial cluster set, and use the average value as the new cluster center to obtain K new cluster centers; the fourth step is to end if the calculated new cluster center is the same as the original cluster center, and obtain K target audio signal sets, otherwise repeat the second step. For example, taking an initial audio signal set as an example, there are 8 audio signals to be classified, namely a, b, c, d, e, f, g, and h, in an initial audio signal set A. The first step is to determine that A is divided into 3 categories, and select b, e, and h in set A as the initial cluster centers; the second step is to calculate the distances from a, c, d, f, and g to b, e, and h respectively, and assign a, c, d, f, and g to the initial cluster centers with the closest distances, and divide the 8 audio signals to be classified in set A into 3 categories with b, e, and h as cluster centers, and obtain 3 initial cluster sets; the third step is to then calculate the average value in each cluster set, and obtain 3 average values ​​of B, E, and H, and use B, E, and H as new cluster centers. The fourth step is to end if the calculated new cluster center is the same as the original cluster center, and obtain 3 target audio signal sets, otherwise repeat the second step. It should be noted that the K-means clustering algorithm is only one embodiment of the present invention, and other clustering algorithms can also be used for cluster analysis, which is not described in detail in the present invention.

[0052] In an embodiment of the present invention, a target audio signal set refers to a category after the initial audio signal set is classified. The target audio signal set is the final category of multiple audio signals to be classified. Each target audio signal set represents a category. Clustering processing is performed on each initial audio signal set to obtain a target audio signal set. Each target audio signal set corresponds to one or more initial audio signal sets.

[0053] In the embodiment of the present invention, each initial audio signal set is divided into one or more target audio signal sets, and multiple initial audio signal sets correspond to multiple target audio signal sets.

[0054] In an embodiment of the present invention, according to the characteristics of each audio signal to be classified, a plurality of audio signals to be classified are roughly classified to obtain a plurality of initial audio signal sets, and then each initial audio signal set is clustered to obtain a plurality of target audio signal sets, wherein each initial audio signal set includes one or more audio signals to be classified; and each initial audio signal set corresponds to one or more target audio signal sets. In this way, by first roughly classifying a plurality of audio signals to be classified, and then finely classifying each initial audio signal set obtained after the rough classification to obtain a target audio signal set, it is possible to finely classify the characteristics of the audio signals to be classified to obtain a variety of rich audio style categories, and further provide users with more audio style categories for users to choose from, solving the problem that users cannot find the desired music style when playing music.

[0055] In the embodiment of the present invention, Figure 2 As shown, Figure 2 is a flow chart of an audio classification method provided by the present invention, wherein the features include: time-frequency features and / or attribute features; step S120: performing initial classification on the multiple audio signals to be classified based on the features of each audio signal to be classified in the multiple audio signals to be classified, and obtaining multiple initial audio signal sets, including:

[0056] Step S121: performing initial classification on the multiple audio signals to be classified based on the time-frequency characteristics and / or attribute characteristics of each of the multiple audio signals to be classified to obtain the multiple initial audio signal sets.

[0057] In the embodiment of the present invention, the multiple audio signals to be classified can be initially classified according to the time-frequency characteristics of each audio signal to be classified; the multiple audio signals to be classified can also be initially classified according to the attribute characteristics of each audio signal to be classified; the multiple audio signals to be classified can also be initially classified according to the time-frequency characteristics and attribute characteristics of each audio signal to be classified, and the embodiment of the present invention does not limit this.

[0058] In addition, both the time-frequency features and the attribute features include multiple features. When classifying based on the features of the audio signal to be classified, the features can be processed by dimensionality reduction, wherein the dimensionality reduction algorithm includes principal component analysis, average pooling and other algorithms, and the embodiments of the present invention do not impose specific restrictions on this.

[0059] In an embodiment of the present invention, based on the time-frequency features and / or attribute features of each of the multiple audio signals to be classified, the multiple audio signals to be classified are initially classified to obtain multiple initial audio signal sets, each of which includes one or more audio signals to be classified. In this way, the time-frequency features can provide a deeper understanding of the information of the audio signals to be classified, and the attribute features can intuitively provide a surface information of the audio signals to be classified. The information of the audio signals to be classified can be fully understood by using the time domain features and / or attribute features, which is conducive to better classification of the audio signals to be classified and obtaining more abundant audio categories. In addition, the use of the dimensionality reduction algorithm is conducive to reducing the amount of calculation in the classification process, thereby improving the classification efficiency.

[0060] In the embodiment of the present invention, Figure 3 As shown, Figure 3 The present invention provides a flowchart of an audio classification method, wherein the time-frequency features include: time domain features and / or frequency domain features; step S121: based on the time-frequency features and / or attribute features of each of the multiple audio signals to be classified, the multiple audio signals to be classified are initially classified to obtain the multiple initial audio signal sets, including:

[0061] Step S1211: Based on the time domain features and / or frequency domain features of each audio signal to be classified, perform a first initial classification on the multiple audio signals to be classified to obtain multiple second audio signal sets; each second audio signal set includes one or more audio signals to be classified.

[0062] Step S1212: Based on the attribute characteristics of each audio signal to be classified, perform a second initial classification on one or more audio signals to be classified in each second audio signal set to obtain the multiple initial audio signal sets; wherein each initial audio signal set corresponds to one or more second audio signal sets; wherein the attribute characteristics include one or more of the following: pitch, duration, timbre, speed and intensity.

[0063] In an embodiment of the present invention, the time-frequency features include time domain features and / or frequency domain features. The time domain features are features obtained by performing time domain analysis on the signal. The time domain analysis methods include but are not limited to short-time energy, wavelet transform, short-time zero-crossing signal, centroid and root mean square, etc.; the frequency domain features are features obtained by performing frequency domain analysis on the signal. The frequency domain analysis methods include but are not limited to power spectrum, inverse spectrum, envelope spectrum, etc.

[0064] Specifically, by performing short-time Fourier transform (STFT) on multiple audio signals to be classified, the analog signal is transformed from the time domain to the frequency domain, and spectrograms of the multiple audio signals to be classified are obtained. The frequency domain features of the audio signals are extracted based on the spectrograms. Other feature extraction algorithms can also be used to extract features, which is not limited in the embodiments of the present invention.

[0065] In the embodiment of the present invention, the attribute characteristics of the audio signal can be intuitively obtained from the audio signal, such as pitch, duration, timbre, speed and intensity.

[0066] It should be understood that the first initial classification can be performed on multiple audio signals to be classified according to the time domain characteristics of each audio signal to be classified; the first initial classification can also be performed on multiple audio signals to be classified according to the frequency domain characteristics of each audio signal to be classified; and the first initial classification can also be performed on multiple audio signals to be classified according to the time domain characteristics and frequency domain characteristics of each audio signal to be classified. The embodiment of the present invention is not limited to this.

[0067] In the embodiment of the present invention, the second audio signal set refers to a category after a plurality of audio signals to be classified are classified. One second audio signal set represents one category, and each second audio signal set includes one audio signal to be classified or a plurality of audio signals to be classified.

[0068] In the embodiment of the present invention, a plurality of audio signals to be classified are subjected to a first initial classification according to the time domain feature and / or frequency domain feature of each audio signal to be classified, so as to obtain a plurality of second audio signal sets.

[0069] In the embodiment of the present invention, the plurality of second audio signal sets refer to a plurality of categories obtained after classifying a plurality of audio signals to be classified, and each second audio signal set includes one or more audio signals to be classified.

[0070] In the embodiment of the present invention, each second audio signal set is divided into one or more initial audio signal sets, and multiple second audio signal sets correspond to multiple initial audio signal sets.

[0071] In an embodiment of the present invention, a first initial classification is performed on a plurality of audio signals to be classified using the time domain features and / or frequency domain features of each audio signal to be classified, and then a second classification is performed on a second audio signal set obtained after the first initial classification according to the attribute features of each audio signal to be classified, so as to obtain a plurality of initial audio signal sets. In this way, through two classifications, one or more audio signals to be classified in the obtained initial audio signal sets are made more identical or similar, which is beneficial to the subsequent classification of audio signals and improves the accuracy of the classification of the audio signals to be classified.

[0072] In the embodiment of the present invention, Figure 4 As shown, Figure 4 is a flow chart of an audio classification method provided by the present invention, step S130: clustering one or more audio signals to be classified in each initial audio signal set to obtain multiple target audio signal sets, including:

[0073] Step S131: determining K cluster centers in each of the initial audio signal sets; K is an integer greater than or equal to 1.

[0074] Step S132: Based on the K cluster centers, determine K cluster sets corresponding to the K cluster centers.

[0075] Step S133: when the similarities between the K cluster sets and the K-1 cluster sets corresponding to the K-1 cluster centers meet a first condition, the K cluster sets are determined as K target audio signal sets.

[0076] In the embodiment of the present invention, K refers to the category obtained after clustering processing. In the clustering processing, K cluster centers need to be initialized and set to divide one or more audio signals to be classified in each initial audio signal set into K categories. Here, K is an integer greater than or equal to 1, and the embodiment of the present invention does not limit the value of K.

[0077] In the embodiment of the present invention, when the K value is initially set, the K value may be set according to the experience of the staff, or the K value may start from 1, which is not specifically limited in the embodiment of the present invention.

[0078] In an embodiment of the present invention, after setting the K value, after dividing one or more audio signals to be classified in each initial audio signal set into K cluster sets, it is necessary to determine whether the K cluster sets are in a convergent state. When the cluster sets are in a convergent state, it means that when each initial audio signal is divided into K cluster sets, the accuracy of cluster classification is the highest.

[0079] In the embodiment of the present invention, the similarities between K cluster sets and K-1 cluster sets corresponding to K-1 cluster centers are calculated. The greater the similarity, the higher the classification accuracy of each initial audio signal set, and the smaller the similarity, the lower the classification accuracy of each initial audio signal set.

[0080] In an embodiment of the present invention, after obtaining multiple initial audio signal sets, clustering processing is performed on one or more audio signals to be classified in each initial audio signal set, and in the process of clustering processing, by continuously changing the number of cluster centers, it is determined that when the similarity between K cluster sets determined by K cluster centers and K-1 cluster sets determined by K-1 cluster centers meets the first condition, the K cluster sets are determined as K target audio signal sets. In this way, when the value of the cluster center is changed, one or more audio signals to be classified in each cluster set can be continuously updated, thereby playing a screening role. When the similarity meets the first condition, it means that the cluster sets converge to achieve the most accurate classification effect.

[0081] In the embodiment of the present invention, step S133: when the similarity between the K cluster sets and the K-1 cluster sets corresponding to the K-1 cluster centers satisfies the first condition, before determining the K cluster sets as the K target audio signal sets, the method further includes:

[0082] Based on the K cluster sets, determining a first intra-class distance; the first intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K cluster sets to the corresponding cluster center;

[0083] Based on the K-1 cluster sets, determining a second intra-class distance; the second intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K-1 cluster sets to the corresponding cluster center;

[0084] Based on the first intra-class distance and the second intra-class distance, it is determined whether the similarity satisfies the first condition.

[0085] In the implementation of the present invention, the distance is used to measure the similarity between two audio signals. The smaller the distance, the greater the similarity, and the larger the distance, the smaller the similarity.

[0086] It should be understood that in the clustering process, the smaller the distance between multiple audio signals to be classified in a target audio signal set, the better, indicating that the similarity of the multiple audio signals to be classified in the target audio signal set is higher. Therefore, it is necessary to calculate the value obtained by adding the squares of the distances from each audio signal to be classified in the K cluster sets to the corresponding cluster center, which is called the first-class intra-distance. The smaller the first-class intra-distance is, the higher the clustering accuracy is.

[0087] In an embodiment of the present invention, the value obtained by adding the squares of the distances from each audio signal to be classified in the K-1 cluster sets determined by the K-1 cluster centers to the corresponding cluster center is called the second intra-class distance. According to the first intra-class distance and the second intra-class distance, it is determined whether the similarity between the K cluster sets and the K-1 cluster sets meets the first condition. If the first condition is met, it means that the classification accuracy is closer. At this time, the K cluster sets are determined as K target audio sets.

[0088] In an embodiment of the present invention, whether the similarity satisfies the first condition is determined based on the first intra-class distance obtained by adding the square of the distance from each audio signal to be classified in the K cluster sets to the corresponding cluster center, and the second intra-class distance obtained by adding the square of the distance from each audio signal to be classified in the K-1 cluster sets to the corresponding cluster center. In this way, by using the first intra-class distance and the second intra-class distance, it is possible to accurately judge whether the K cluster sets determined by the K cluster centers are in a convergence state and whether the most accurate classification effect is achieved, thereby improving the accuracy of clustering.

[0089] In the embodiment of the present invention, determining whether the similarity satisfies the first condition based on the first intra-category distance and the second intra-category distance includes:

[0090] Based on the first intra-class distance, determining a first ratio; wherein the first ratio is a ratio of the first inter-class distance to the first intra-class distance, and the first inter-class distance is obtained by adding the sum of the squares of the distances between the K cluster centers and the centers of the corresponding initial audio signal sets;

[0091] Determine a second ratio based on the second intra-class distance; wherein the second ratio is a ratio of the second inter-class distance to the second intra-class distance, and the second inter-class distance is obtained by adding the sum of the squares of the distances between the K-1 cluster centers and the centers of the corresponding initial audio signal sets;

[0092] Based on a difference between the first ratio and the second ratio, it is determined whether the similarity satisfies the first condition.

[0093] In an embodiment of the present invention, the first inter-class distance is obtained by adding the sum of the squares of the distances from the center of each cluster set to the center of the corresponding initial audio signal set after each initial audio signal set is divided into K cluster sets, and the second inter-class distance is obtained by adding the sum of the squares of the distances from the center of each cluster set to the center of the corresponding initial audio signal set after each initial audio signal set is divided into K-1 cluster sets.

[0094] It should be understood that when clustering analysis is performed on one or more audio signals to be classified in the initial audio signal, the smaller the intra-class distance of each class after classification, the more similar the multiple audio signals to be classified within the class are, and the larger the clustering between classes, the greater the similarity between classes. Therefore, it is necessary to ensure that the intra-class distance is smaller and the inter-class distance is larger. The larger the ratio of the inter-class clustering to the intra-class distance is, the higher the classification accuracy of the obtained clustering set is.

[0095] In an embodiment of the present invention, after determining the distance within the first class and the distance within the second class, a first ratio is determined according to the distance within the first class, and a second ratio is determined according to the clustering within the second class. According to the difference between the first ratio and the second ratio, it is determined whether the similarity satisfies the first condition. Thus, when the difference between the first ratio and the second ratio is less than the preset value, it indicates that the cluster set obtained at this time is in a convergence state, the similarity satisfies the first condition, and the cluster analysis achieves the optimal result. By utilizing the inter-class distance and the intra-class distance, the accuracy of the classification of the audio signals to be classified in the cluster set is determined, the accuracy of clustering is improved, and the accuracy of audio classification is further improved. According to the difference between the first ratio and the second ratio, it is determined whether the similarity satisfies the first condition, and it is ensured that the cluster set is in a convergence state, and the accuracy of cluster classification is further improved.

[0096] For example, Figure 5 As shown, Figure 5 The flowchart of the clustering process and updating the clustering set is shown in FIG. 1 . After obtaining the initial audio signal set, the characteristic data of multiple audio signals to be classified in the initial audio signal set are standardized, and then K cluster centers are initialized. The elbow rule and the clustering evaluation index (Calinski-Harabasz, CH) score can be used to determine the K value. The audio signals to be classified (referring to Figure 5 The music in the cluster) is assigned to the nearest audio database, and the K value is determined to be convergent. If it is not converged, the K value is re-determined and the music clustering database is updated. If it converges, the target audio signal set (referring to Figure 5 Clustering with labels in music).

[0097] In the embodiment of the present invention, Figure 6 As shown, Figure 6 is a flow chart of an audio classification method provided by the present invention. In step S110, before extracting features of a plurality of audio signals to be classified, the method further includes:

[0098] Step S101: Acquire the plurality of initial audio signals to be classified.

[0099] Step S102: performing standardization processing on each initial audio signal to be classified to obtain each audio signal to be classified; the standardization processing includes one or more of the following: averaging multi-channel data, removing silent segment signals of the audio, and normalizing the signal.

[0100] In the embodiment of the present invention, multi-channel data generally refers to multiple independent audio tracks. Multiple independent sound sources can help provide the user with a surround sound feeling. By averaging the multi-channel data, the audio received by the user in each direction can be basically consistent without causing too much difference.

[0101] Removing the silent segment signal of the audio refers to removing the silent segment in the initial audio signal so that the user can always hear the audio.

[0102] Signal normalization refers to scaling the eigenvalues ​​of an audio signal to the interval [0, 1] to avoid large deviations between data, resulting in incomparability and accelerating convergence.

[0103] In an embodiment of the present invention, after obtaining multiple initial audio signals to be classified, multi-channel data is averaged for each initial audio signal to be classified, and the silent segment signal of the audio and one or more standardization processes in the normalization process of the signal are removed to avoid the existence of multiple signals with large differences in the initial audio signals to be classified, which affects the accuracy of subsequent classification of the audio signals.

[0104] The audio classification method provided by the embodiment of the present invention is described in detail below in conjunction with specific application scenarios.

[0105] It should be understood that this application scenario is an audio classification method for a vehicle audio system, and the music signal in this application scenario is the audio signal in the present invention.

[0106] refer to Figure 7 As shown, the application of the audio feature classification method of car audio style includes the following steps:

[0107] Step 1: Collect original music samples and construct a music dataset based on the sources of car music: online music in the car, CD music, music imported from a USB flash drive, music in radio channels, online music in mobile phones, music in WeChat or phone voice, and audio information in navigation.

[0108] It should be understood that the collection of original music samples in this application scenario may be the acquisition of multiple initial audio signals to be classified in the present invention.

[0109] Step 2: Preprocess the music samples: average multi-channel data, remove silent signals from the audio, and normalize the signals to remove details that are not related to the music.

[0110] It should be understood that the preprocessing of the music samples in this application scenario may be the standardized processing in the present invention.

[0111] Step 3: Analyze the music samples after removing interference items, and select time domain analysis method or frequency domain analysis method according to the music audio characteristics.

[0112] It should be understood that in this application scenario, according to the music audio features, the time domain analysis method or the frequency domain analysis method is selected, which can be based on the time domain features and / or frequency domain features of each audio signal to be classified in the present invention, and the multiple audio signals to be classified are subjected to a first initial classification to obtain multiple second audio signal sets; each second audio signal set includes one or more audio signals to be classified, wherein the reference Figure 7 As shown, the frequency domain analysis methods include: power spectrum, inverse spectrum, envelope spectrum, etc., and the time domain analysis methods include: short-time energy, wavelet change, short-time zero-crossing rate, centroid, root mean square, etc.

[0113] Step 4: Classify according to the music audio features: pitch, duration, timbre, speed and intensity, and mark the music feature labels to form a music audio feature dataset.

[0114] It should be understood that based on the attribute characteristics of each audio signal to be classified, a second initial classification is performed on one or more audio signals to be classified in each second audio signal set to obtain the multiple initial audio signal sets; wherein, each initial audio signal set corresponds to one or more second audio signal sets, and the music feature labels in this application scenario can be the initial audio signal set in the present invention, each initial audio signal set represents a music feature label, and multiple music feature labels form a music audio feature data set, reference Figure 7 As shown, in the audio classification part, the attribute features include pitch, duration, timbre, speed and intensity. Among them, the methods for extracting pitch include short-time zero-crossing rate, short-time energy, etc.; the methods for extracting duration include wavelet transform, short-time zero-crossing rate, power spectrum, etc.; the methods for extracting timbre include short-time zero-crossing rate, Mel-scale Frequency Cepstral Coefficients (MFCCS); the methods for extracting speed include wavelet analysis, short-time zero-crossing rate, Mel-scale Frequency Cepstral Coefficients (MFCCS), etc.; the methods for extracting intensity include short-time energy, centroid and root mean square. The classification methods include: Back Propagation (BP) neural network, K-value, etc.

[0115] Step 5: Use data clustering analysis method to process the music classified according to the characteristics of the signal, and use the elbow rule and CH score to determine the K value.

[0116] It should be understood that step five may be determining K cluster centers in each initial audio signal set in the present invention; K is an integer greater than or equal to 1; and based on the K cluster centers, determining K cluster sets corresponding to the K cluster centers.

[0117] Step 6: According to the K value, assign each music frequency feature to the nearest audio database, update the identified music clustering database, and combine the convergence of the K value.

[0118] It should be understood that step five is to screen and update the multiple audio signals to be classified in the audio cluster set, which can be the case in the present invention where the similarity between the K cluster sets and the K-1 cluster sets corresponding to the K-1 cluster centers meets the first condition, and the K cluster sets are determined as K target audio signal sets.

[0119] Step 7: Output the clustered labeled music. When the K value does not converge, repeat steps 5-6 until all the music is classified according to the audio features.

[0120] It should be understood that when the K value does not converge, it means that the accuracy of the cluster set determined by the K value is not the highest, and the cluster center needs to be determined and the multiple audio signals to be classified in the cluster set determined by the K value are updated.

[0121] In this application scenario, the original music signal is obtained, and the original music signal is preprocessed by averaging multi-channel data, removing the silent segment signal of the audio, and normalizing the signal; then the preprocessed music signal is analyzed for time-frequency characteristics to obtain time domain characteristics and / or frequency domain characteristics, and multiple music signals are classified according to the time domain characteristics and / or frequency domain characteristics to obtain multiple second music signal sets, each of which includes one or more original music signals, and then the attribute characteristics of the music signal are obtained: pitch, duration, timbre, speed and intensity, and each second music signal set is classified according to the attribute characteristics to obtain A plurality of initial music signal sets, each initial music signal set corresponds to one or more second music signal sets, and then one or more music signals in each initial music signal set are clustered to obtain a plurality of third music signal sets; then the classification accuracy of the third music signal sets is judged, if the K value of the third music signal set converges, the target music signal set determined by the third music signal set is stored in different music libraries, if the K value of the third music set does not converge, the music signals in the third music signal set are updated until the clustering set converges, and the target music signal set is determined to be stored in different music libraries.

[0122] In the embodiment of the present invention, the machine learning data processing module includes the preprocessing of music signals, the audio data algorithm analysis system, the audio player and the speaker. The preprocessing of audio data, the operation of the audio feature analysis system, the operation of the audio feature analysis system, and the application of the clustering calculation method are all operated in the vehicle terminal control system.

[0123] In an embodiment of the present invention, after the audio signal to be classified is classified using the audio classification method provided by the embodiment of the present invention, it can be stored in an audio classification database, so that the characteristic values ​​in the audio signal to be classified can be directly matched with the characteristic values ​​in the audio classification database, and the audio to be classified can be directly classified and identified, and relevant calculations can be performed through machine learning. Based on the self-learning algorithm, music with similar styles can be recommended to users, so that drivers or passengers can enjoy the psychological pleasure brought by music while enjoying the fun of driving.

[0124] The embodiment of the present invention also provides an audio classification device, referring to Figure 8 As shown, the audio classification device includes:

[0125] Extraction module 801: used to extract features of multiple audio signals to be classified;

[0126] The first classification module 802 is used to perform initial classification on the multiple audio signals to be classified based on the characteristics of each audio signal to be classified in the multiple audio signals to be classified, so as to obtain multiple initial audio signal sets; each initial audio signal set includes one or more audio signals to be classified;

[0127] The second classification module 803 is configured to perform clustering processing on one or more audio signals to be classified in each initial audio signal set to obtain multiple target audio signal sets; wherein each initial audio signal set corresponds to one or more target audio signal sets.

[0128] In some embodiments, the first classification module is also used for the features including: time-frequency features and / or attribute features; based on the time-frequency features and / or attribute features of each audio signal to be classified in the multiple audio signals to be classified, initially classify the multiple audio signals to be classified to obtain the multiple initial audio signal sets.

[0129] In some embodiments, the first classification module is also used for the time-frequency features including: time domain features and / or frequency domain features; based on the time domain features and / or frequency domain features of each audio signal to be classified, the multiple audio signals to be classified are subjected to a first initial classification to obtain multiple second audio signal sets; each second audio signal set includes one or more audio signals to be classified; based on the attribute features of each audio signal to be classified, one or more audio signals to be classified in each second audio signal set are subjected to a second initial classification to obtain the multiple initial audio signal sets; wherein, each initial audio signal set corresponds to one or more second audio signal sets; wherein, the attribute features include one or more of the following: pitch, duration, timbre, speed and intensity.

[0130] In some embodiments, the second classification module is also used to determine K cluster centers in each initial audio signal set; K is an integer greater than or equal to 1; based on the K cluster centers, K cluster sets corresponding to the K cluster centers are determined; when the similarity between the K cluster sets and the K-1 cluster sets corresponding to the K-1 cluster centers meets the first condition, the K cluster sets are determined as K target audio signal sets.

[0131] In some embodiments, the second classification module is further used to determine the first intra-class distance based on the K cluster sets; the first intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K cluster sets to the corresponding cluster center; based on the K-1 cluster sets, determine the second intra-class distance; the second intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K-1 cluster sets to the corresponding cluster center; based on the first intra-class distance and the second intra-class distance, determine whether the similarity meets the first condition.

[0132] In some embodiments, the second classification module is also used to determine a first ratio based on the first intra-class distance; wherein the first ratio is the ratio of the first inter-class distance to the first intra-class distance, and the first inter-class distance is obtained by adding the sum of the squares of the distances between the K cluster centers and the centers of the corresponding initial audio signal set; based on the second intra-class distance, determine a second ratio; wherein the second ratio is the ratio of the second inter-class distance to the second intra-class distance, and the second inter-class distance is obtained by adding the sum of the squares of the distances between the K-1 cluster centers and the centers of the corresponding initial audio signal set; based on the difference between the first ratio and the second ratio, determine whether the similarity meets the first condition.

[0133] In some embodiments, the extraction module is also used to obtain the multiple initial audio signals to be classified; perform standardization processing on each initial audio signal to be classified to obtain each audio signal to be classified; the standardization processing includes one or more of the following: averaging multi-channel data, removing silent segment signals of the audio, and normalizing the signal.

[0134] Of course, in practical applications, Fig. 9 As shown, the various components in the electronic device are coupled together via a bus system 902. It is understood that the bus system 902 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 902 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Fig. 9 The various buses are collectively labeled bus system 902 .

[0135] It can be understood that the memory in this embodiment can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memories.

[0136] The method disclosed in the above embodiment of the present invention can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or instructions in software form. The above processor may be a general-purpose processor, a DSP, or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present invention. A general-purpose processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0137] The embodiment of the present invention further provides a computer storage medium, specifically a computer readable storage medium, on which computer instructions are stored. As a first implementation, when the computer storage medium is located in an electronic device, the computer instructions are executed by a processor to implement any step of the above-mentioned audio classification method of the embodiment of the present invention.

[0138] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0139] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0140] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or at least two units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0141] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium, which, when executed, executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, disks or optical disks.

[0142] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0143] It should be noted that the technical solutions described in the embodiments of the present invention can be combined arbitrarily without conflict.

[0144] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An audio classification method, characterized in that: The method comprises: Extracting features of multiple audio signals to be classified; Based on the feature of each of the multiple audio signals to be classified, initially classify the multiple audio signals to be classified to obtain multiple initial audio signal sets; each initial audio signal set includes one or more audio signals to be classified; Clustering is performed on one or more audio signals to be classified in each of the initial audio signal sets to obtain multiple target audio signal sets; wherein each of the initial audio signal sets corresponds to one or more target audio signal sets.

2. The method according to claim 1, characterized in that The features include: time-frequency features and / or attribute features; the multiple audio signals to be classified are initially classified based on the features of each audio signal to be classified in the multiple audio signals to be classified to obtain multiple initial audio signal sets, including: Based on the time-frequency characteristics and / or attribute characteristics of each of the multiple audio signals to be classified, the multiple audio signals to be classified are initially classified to obtain the multiple initial audio signal sets.

3. The method according to claim 2, characterized in that The time-frequency features include: time domain features and / or frequency domain features; the initial classification of the multiple audio signals to be classified based on the time-frequency features and / or attribute features of each audio signal to be classified in the multiple audio signals to be classified to obtain the multiple initial audio signal sets includes: Based on the time domain features and / or frequency domain features of each audio signal to be classified, performing a first initial classification on the multiple audio signals to be classified to obtain multiple second audio signal sets; each second audio signal set includes one or more audio signals to be classified; Based on the attribute characteristics of each audio signal to be classified, a second initial classification is performed on one or more audio signals to be classified in each second audio signal set to obtain the multiple initial audio signal sets; wherein each initial audio signal set corresponds to one or more second audio signal sets; wherein the attribute characteristics include one or more of the following: pitch, duration, timbre, speed and intensity.

4. The method according to any one of claims 1 to 3, characterized in that: The clustering process is performed on one or more audio signals to be classified in each of the initial audio signal sets to obtain multiple target audio signal sets, including: Determine K cluster centers in each of the initial audio signal sets; K is an integer greater than or equal to 1; Based on the K cluster centers, determining K cluster sets corresponding to the K cluster centers; When the similarities between the K cluster sets and the K-1 cluster sets corresponding to the K-1 cluster centers meet a first condition, the K cluster sets are determined as K target audio signal sets.

5. The method according to claim 4, characterized in that When the similarities between the K cluster sets and the K-1 cluster sets corresponding to the K-1 cluster centers meet the first condition, before determining the K cluster sets as the K target audio signal sets, the method further includes: Based on the K cluster sets, determining a first intra-class distance; the first intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K cluster sets to the corresponding cluster center; Based on the K-1 cluster sets, determining a second intra-class distance; the second intra-class distance is obtained by adding the square sum of the distances from each audio signal to be classified in the K-1 cluster sets to the corresponding cluster center; Based on the first intra-class distance and the second intra-class distance, it is determined whether the similarity satisfies the first condition.

6. The method according to claim 5, characterized in that The determining, based on the first intra-class distance and the second intra-class distance, whether the similarity satisfies the first condition includes: Based on the first intra-class distance, determining a first ratio; wherein the first ratio is a ratio of the first inter-class distance to the first intra-class distance, and the first inter-class distance is obtained by adding the sum of the squares of the distances between the K cluster centers and the centers of the corresponding initial audio signal sets; Determine a second ratio based on the second intra-class distance; wherein the second ratio is a ratio of the second inter-class distance to the second intra-class distance, and the second inter-class distance is obtained by adding the sum of the squares of the distances between the K-1 cluster centers and the centers of the corresponding initial audio signal sets; Based on a difference between the first ratio and the second ratio, it is determined whether the similarity satisfies the first condition.

7. The method according to any one of claims 1 to 3, characterized in that: Before extracting the features of the plurality of audio signals to be classified, the method further comprises: Acquire the multiple initial audio signals to be classified; Each initial audio signal to be classified is subjected to standardization processing to obtain each audio signal to be classified; the standardization processing includes one or more of the following: averaging multi-channel data, removing silent segment signals of the audio, and normalizing the signal.

8. An audio classification device, characterized in that: The device comprises: An extraction module, used for extracting features of multiple audio signals to be classified; A first classification module is used to perform initial classification on the multiple audio signals to be classified based on the characteristics of each of the multiple audio signals to be classified to obtain multiple initial audio signal sets; each initial audio signal set includes one or more audio signals to be classified; The second classification module is used to perform clustering processing on one or more audio signals to be classified in each initial audio signal set to obtain multiple target audio signal sets; wherein each initial audio signal set corresponds to one or more target audio signal sets.

9. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the audio classification method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the audio classification method according to any one of claims 1 to 7 is implemented.

Images