Method and device for extracting music characteristics

Abstract

The invention discloses a method and a device for extracting music characteristics. The method comprises the following steps: segmenting a received audio signal to generate at least two segmented audio signals; carrying out Fourier transform on each segmented audio signal to obtain a frequency domain signal of each segmented audio signal; calculating a frequency mass center of each segmented audio signal according to the frequency domain signal of each segmented audio signal and a corresponding frequency of the frequency domain signal, and taking the frequency mass center as a music characteristic; calculating bandwidth of each segmented audio signal according to the frequency mass center and the frequency domain signal and the corresponding frequency of the frequency domain signal of each segmented audio signal, and taking the bandwidth as a music characteristic. Through the method and the device, the received audio signal is segmented; the mass center and the bandwidth of each segmented audio signal are calculated according to all the frequency information meeting the condition in each segmented audio signal; and the mass center and the bandwidth serve as the music characteristics, thereby reducing the influence of the environment on the audio signal characteristics and improving the recognition rate of a system.

Description

technical field

[0001] The invention relates to signal processing technology, in particular to a method and device for extracting music features. Background technique

[0002] CMI (Contend-based Music Identification, content-based music identification) is currently a popular application on smartphones. Its application scenario is: when a user hears a piece of music he likes but does not know the title of the song, he can record a few seconds of music clips through his mobile phone, and then the background system will find various information about the music through search technology and feed it back to user. In order to realize this function, the first task is to extract appropriate music features from a large number of training music databases, and establish a training set feature index database as the basis for subsequent feature matching of the music segments to be tested.

[0003] Feature extraction is an important part of the CMI system. Most of the features used in t...

Examples

Embodiment 1

[0022] figure 2 It is a flow chart of the method for extracting music features provided by the first embodiment of the present invention. The execution subject of this embodiment can be a feature extraction unit, and the feature extraction unit can also be called a device for extracting music features. It is composed of hardware and / or Software implementation, can be configured in the local client, can also be configured in the server in the network, not specifically limited here, the method provided in this embodiment specifically includes the following steps:

[0023] Step 201: Segment the received audio signal to generate at least two segmented audio signals.

[0024] In this solution, the audio signal can be from any source, for example, the user records the audio signal himself, or receives the audio signal acquired. Preferably, the received audio signal may be segmented into at least two segments of audio signals of equal length, or the received audio signal may be seg...

no. 2 example

[0043] On the basis of the above-mentioned embodiments, this embodiment further adds a step of calculating each segment according to the music feature of each segmented audio signal after acquiring the music feature of each segmented audio signal. Differential features of audio signals, as musical features. The steps may be performed after step 203 and before step 204, or after step 204, or both after step 203 and after step 204, which are not specifically limited here.

[0044] For example, after the frequency centroid of each segment audio signal is obtained, the frequency centroid of the current segment audio signal can be subtracted from the frequency centroid of the previous segment audio signal, as the frequency centroid difference feature of the current segment audio signal, used to describe the frequency The change law of the centroid; when the bandwidth of each segment audio signal is obtained, the bandwidth of the current segment audio signal can be subtracted from t...

no. 3 example

[0047] In this embodiment, on the basis of the above-mentioned embodiments, a further step is added: after the music features are obtained, the extracted music features are concatenated into a multi-dimensional vector, and the multi-dimensional vector is subjected to dimensionality reduction processing.

[0048] After obtaining a certain music feature of a certain segment of audio signal, usually the feature will not be used alone, but several music features will be combined to form a high-dimensional feature vector, so as to describe a segment of audio signal more accurately. The dimensionality of the newly constructed feature vectors may be relatively high. On the one hand, the dimensionality reduction technology can reduce the dimensionality of the feature vectors, reduce the amount of calculations for subsequent establishment of feature indexes and feature matching, and on the other hand, it can also reduce the relationship between the dimensions of the feature vectors. The...

technical field

[0001] The invention relates to signal processing technology, in particular to a method and device for extracting music features. Background technique

[0002] CMI (Contend-based Music Identification, content-based music identification) is currently a popular application on smartphones. Its application scenario is: when a user hears a piece of music he likes but does not know the title of the song, he can record a few seconds of music clips through his mobile phone, and then the background system will find various information about the music through search technology and feed it back to user. In order to realize this function, the first task is to extract appropriate music features from a large number of training music databases, and establish a training set feature index database as the basis for subsequent feature matching of the music segments to be tested.

[0003] Feature extraction is an important part of the CMI system. Most of the features used in t...

Examples

Embodiment 1

[0022] figure 2 It is a flow chart of the method for extracting music features provided by the first embodiment of the present invention. The execution subject of this embodiment can be a feature extraction unit, and the feature extraction unit can also be called a device for extracting music features. It is composed of hardware and / or Software implementation, can be configured in the local client, can also be configured in the server in the network, not specifically limited here, the method provided in this embodiment specifically includes the following steps:

[0023] Step 201: Segment the received audio signal to generate at least two segmented audio signals.

[0024] In this solution, the audio signal can be from any source, for example, the user records the audio signal himself, or receives the audio signal acquired. Preferably, the received audio signal may be segmented into at least two segments of audio signals of equal length, or the received audio signal may be seg...

no. 2 example

[0043] On the basis of the above-mentioned embodiments, this embodiment further adds a step of calculating each segment according to the music feature of each segmented audio signal after acquiring the music feature of each segmented audio signal. Differential features of audio signals, as musical features. The steps may be performed after step 203 and before step 204, or after step 204, or both after step 203 and after step 204, which are not specifically limited here.

[0044] For example, after the frequency centroid of each segment audio signal is obtained, the frequency centroid of the current segment audio signal can be subtracted from the frequency centroid of the previous segment audio signal, as the frequency centroid difference feature of the current segment audio signal, used to describe the frequency The change law of the centroid; when the bandwidth of each segment audio signal is obtained, the bandwidth of the current segment audio signal can be subtracted from t...

no. 3 example

[0047] In this embodiment, on the basis of the above-mentioned embodiments, a further step is added: after the music features are obtained, the extracted music features are concatenated into a multi-dimensional vector, and the multi-dimensional vector is subjected to dimensionality reduction processing.

[0048] After obtaining a certain music feature of a certain segment of audio signal, usually the feature will not be used alone, but several music features will be combined to form a high-dimensional feature vector, so as to describe a segment of audio signal more accurately. The dimensionality of the newly constructed feature vectors may be relatively high. On the one hand, the dimensionality reduction technology can reduce the dimensionality of the feature vectors, reduce the amount of calculations for subsequent establishment of feature indexes and feature matching, and on the other hand, it can also reduce the relationship between the dimensions of the feature vectors. The...

Images