A method and system for identifying five-tone music mode

Abstract

The present invention belongs to the technical field of mode recognition, and specifically is a five-tone music mode recognition method and system. The method of the present invention comprises: obtaining music audio to be recognized; extracting the chromaticity features of the music audio to be recognized; constructing and training a mode recognition model; inputting the chromaticity features of the music audio to be recognized into the trained mode recognition model, so as to obtain the mode of the music audio to be recognized; the system of the present invention comprises 4 modules accordingly; the present invention directly recognizes the mode of the music audio, which can improve the recognition accuracy and the recognition reliability.

Description

Technical Field

[0001] The invention belongs to the technical field of mode recognition, and in particular relates to a method and system for recognizing pentatonic music modes. Background Art

[0002] A mode is a musical structure that organizes a number of musical notes of different pitches around a stable central tone, organized in a certain interval relationship to form an organic system. The central tone of the mode is used as the starting and ending tone, and the other tones are arranged in the order of pitch in the form of a scale, which is called a modal scale.

[0003] At present, there are few studies on the recognition of Chinese pentatonic music modes. The current methods for recognizing pentatonic music modes include: Chinese folk music pentatonic mode recognition algorithm based on decision tree, Chinese folk music mode recognition method based on template matching, etc. The above methods are for recognition of music scores and midi files, but due to the small number of samples, the recognition accuracy is not high and the reliability needs to be improved. Summary of the invention

[0004] The invention provides a method and system for recognizing pentatonic music modes with high recognition accuracy and good reliability.

[0005] The five-tone music mode recognition method provided by the present invention comprises the following specific steps:

[0006] (1) Obtain the music audio to be identified;

[0007] (2) extracting the chromaticity features of the music audio to be identified;

[0008] (3) Build and train a mode recognition model;

[0009] (4) The chromaticity features of the music audio to be identified are input into a trained mode recognition model to obtain the mode of the music audio to be identified.

[0010] In the present invention, the step (2) of extracting the chromaticity features of the music audio to be identified specifically includes:

[0011] (2.1) performing a preprocessing operation on the music audio to be identified to obtain a frequency spectrum of the music audio to be identified;

[0012] (2.2) Converting the spectrum of the music audio to be identified into a chromatogram, wherein the chromatogram of the music audio to be identified is the chromaticity feature of the music audio to be identified.

[0013] Furthermore, the preprocessing operation of the music audio to be recognized in step (2.1) specifically includes:

[0014] (2.1.1) performing pre-emphasis processing on the music audio to be identified to obtain emphasized music audio;

[0015] (2.1.2) performing frame processing on the emphasized music audio to obtain framed music audio;

[0016] (2.1.3) performing windowing processing on the framed music audio to obtain windowed music audio;

[0017] (2.1.4) Performing a short-time Fourier transform on the windowed music audio to obtain a spectrum diagram of the music audio to be identified.

[0018] Optionally, the emphasized music audio in step (2.1.2) is subjected to frame processing, specifically comprising:

[0019] The pre-emphasized music audio is divided into frames with 512 sampling points as one frame to obtain the divided music audio; the time length of each frame in the divided music audio is 11.61ms, and the overlap rate between frames is 50%.

[0020] Furthermore, the step (2.2) of converting the spectrum of the music audio to be identified into a chromatogram specifically includes:

[0021] (2.2.1) removing overtones from the spectrum of the music audio to be identified, to obtain a spectrum after removal;

[0022] (2.2.2) tuning the removed spectrum to obtain a tuned spectrum;

[0023] (2.2.3) converting the tuned spectrogram into an initial chromatogram using a loudness-based method;

[0024] (2.2.4) The octaves of the initial chromatogram are merged, and the merged chromatogram is normalized to obtain the chromatogram of the music audio to be identified.

[0025] In the present invention, the mode recognition model is constructed in step (3), and the mode recognition model specifically includes a first long short-term memory model layer (LSTM), a Dropout layer, a second long short-term memory model layer (LSTM), a fully connected layer and a softmax function connected in sequence.

[0026] In the present invention, the training of the mode recognition model in step (3) specifically includes:

[0027] (3.1) obtaining a training set; the training set includes training music audio and corresponding annotated modes;

[0028] (3.2) extracting the chromaticity features of the training music audio;

[0029] (3.3) The chromaticity features of the training music audio and the corresponding annotated modes are input into a machine learning model for training, and the trained machine learning model is determined as a mode recognition model.

[0030] Furthermore, the step (3.2) of extracting the chromaticity features of the training music audio specifically includes:

[0031] (3.2.1) performing data enhancement processing on the training music audio to obtain enhanced training music audio; the data enhancement processing includes speed conversion and pitch conversion;

[0032] (3.2.2) performing a preprocessing operation on the enhanced training music audio to obtain a frequency spectrum of the training music audio;

[0033] (3.2.3) Converting the spectrum of the training music audio into a chromatogram, wherein the chromatogram of the training music audio is a chromaticity feature of the training music audio.

[0034] Optionally, the step (3.2.1) of performing data enhancement processing on the training music audio to obtain enhanced training music audio specifically includes:

[0035] (3.2.1.1) transforming the speed of the training music audio into 1.1, 1.2, 0.9 and 0.8 times the speed respectively, to obtain speed-changed music audio; the speed-changed music audio includes the music audio before the speed change and the music audio after the speed change;

[0036] (3.2.1.2) raising and lowering the pitch of the speed-changed music audio by 1-3 semitones respectively to obtain pitch-changed music audio; the pitch-changed music audio includes the music audio before the pitch change and the music audio after the pitch change;

[0037] (3.2.1.3) Determine the pitch-changed music audio as enhanced training music audio.

[0038] The present invention also provides a five-tone music mode recognition system, comprising:

[0039] An audio acquisition module, used to acquire the music audio to be recognized;

[0040] A chromaticity feature extraction module, used to extract the chromaticity features of the music audio to be identified;

[0041] Mode recognition model module;

[0042] The mode recognition model training and recognition module is used to input the chromaticity features of the music audio to be recognized into the mode recognition model to obtain the mode of the music audio to be recognized; the mode recognition model is obtained by training the machine learning model using a training set.

[0043] The above four modules correspond to the operations of the four steps of the five-tone music mode recognition method of the present invention.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention directly performs mode recognition on music audio, solves the problem of low recognition accuracy and reliability due to too small a number of samples when recognizing music scores and midi files, and greatly improves recognition accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0047] Figure 1 The present invention provides a flowchart of a method for identifying five-tone music modes.

[0048] Figure 2 A flow chart of a method for determining a mode recognition model provided by an embodiment of the present invention.

[0049] Figure 3 A flowchart of data enhancement and preprocessing operations for training music audio provided by an embodiment of the present invention.

[0050] Figure 4 A flow chart of converting a spectrum diagram into a chromatogram provided in an embodiment of the present invention.

[0051] Figure 5 A structural diagram of a recurrent neural network provided in an embodiment of the present invention.

[0052] Figure 6 A structural diagram of a machine learning model provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0053] 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.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] At present, in the study of traditional Chinese five-tone therapy, a large number of traditional music audios with marked modes are needed. The concept of traditional Chinese five-element music modes is vague, and the mode changes frequently. It is time-consuming and laborious to determine the mode of a piece of music, and it is difficult to standardize. In addition, there are many different versions of the same piece of music, and different versions lead to inconsistent modes. For the Shang mode and the Jiao mode, it is even more difficult to find enough tracks because they end on unstable notes. Therefore, there is an urgent need for a mode recognition method for audio.

[0056] Figure 1 Flow chart of the five-tone music mode recognition method provided by the embodiment of the present invention. Figure 1 The five-tone music mode recognition method of this embodiment directly performs mode recognition on music audio. The method includes:

[0057] Step 101: Obtain music audio to be recognized.

[0058] Step 102: extracting the chromaticity features of the music audio to be identified.

[0059] The step 102 specifically includes:

[0060] (1) Perform preprocessing operations on the music audio to be identified to obtain a spectrum diagram of the music audio to be identified. Specifically:

[0061] The music audio to be identified is pre-emphasized to obtain the emphasized music audio. The music audio after the emphasis is framed to obtain the framed music audio. The music audio after the frame is windowed to obtain the windowed music audio. The music audio after the windowing is short-time Fourier transformed to obtain the spectrum of the music audio to be identified.

[0062] Pre-emphasis processing: Pre-emphasis is a signal processing method that compensates for the high-frequency components of the input signal at the transmitting end. During the transmission process, there is always a certain degree of loss, so the signal needs to be compensated at the receiving end. The idea of ​​pre-emphasis technology is to enhance the high-frequency components of the signal at the beginning of the transmission line to compensate for the excessive attenuation of the high-frequency components during the transmission process. Pre-emphasis has no effect on noise, so it effectively improves the output signal-to-noise ratio.

[0063] Frame processing: When processing audio signals, Fourier transform is generally used to convert the speech signal from the time domain to the frequency domain. The frequency of the audio signal is constantly changing, that is, "non-periodic", and the change is stable in a short time, that is, it has "short-term stability" (the audio signal can be considered to be approximately unchanged within 10-30ms), so the long speech can be truncated into short segments for "short-term analysis". At this time, it can be considered that this short-term speech is periodic. The music audio used in this embodiment is 44100Hz, that is, the music audio after emphasis is 44100Hz, and the music audio after emphasis is framed in a way of 512 sampling points as one frame to obtain the music audio after framing; the time length of each frame in the music audio after framing is 11.61ms, and the overlap rate between frames is 50%, that is, 256 sampling points overlap between frames.

[0064] Windowing: After the above framing, a windowing function needs to be added to the frame. The purpose of windowing is to intercept a small segment of audio for periodic extension and Fourier transform. Different window functions have different spectrum leakages, which can be selected according to data and task requirements. Therefore, a window function is used to perform windowing on the music audio after framing to obtain the windowed music audio; the window function is a rectangular window, a Hanning window or a Hamming window.

[0065] ①Rectangular window

[0066] The advantage of this window is that the main lobe is relatively concentrated, but the disadvantage is that the side lobes are relatively high and have negative side lobes, which lead to high-frequency interference and leakage in the conversion, and even the occurrence of negative spectrum phenomenon.

[0067]

[0068] ②Hanning Window

[0069] The main lobe of the Hanning window is widened and lowered, while the side lobes are significantly reduced. From the perspective of reducing leakage, the Hanning window is better than the rectangular window. However, the widening of the main lobe of the Hanning window is equivalent to widening the analysis bandwidth and reducing the frequency resolution.

[0070]

[0071] ③Hamming Window

[0072] The Hamming window and the Hanning window are both cosine windows, but with different weighting coefficients. The weighted coefficients of the Hamming window can make the side lobes smaller.

[0073]

[0074] Analysis shows that the first side lobe attenuation of the Hamming window is 42 dB, and the attenuation speed of the Hamming window side lobe is 20 dB / (10 oct), which is slower than the attenuation speed of the Hanning window. In this embodiment, the Hamming window is selected for windowing.

[0075] Short-time Fourier transform (STFT): Before the signal is Fourier transformed, it is multiplied by a time-limited window function h(t), and it is assumed that the non-stationary signal is stationary within the short time interval of the analysis window. By moving the window function h(t) on the time axis, the signal is analyzed segment by segment to obtain a set of local "spectra" of the signal. The short-time Fourier transform of the signal x(t) is defined as:

[0076]

[0077] After STFT, the spectrum of the music audio to be recognized will be obtained.

[0078] (2) After obtaining the spectrum of the music audio to be identified, it is necessary to perform operations such as background noise removal, tuning, pitch extraction, and octave normalization, so as to convert the spectrum of the music audio to be identified into a chromatogram. The chromatogram of the music audio to be identified is the chromaticity feature of the music audio to be identified. Specifically, the background noise in the spectrum of the music audio to be identified is removed to obtain the spectrum after removal. The spectrum after removal is tuned to obtain the tuned spectrum. The tuned spectrum is converted into an initial chromatogram using a loudness-based method. The octaves of the initial chromatogram are merged, and the merged chromatogram is normalized to obtain the chromatogram of the music audio to be identified.

[0079] Removing background noise: The percussion elements in music can cause background noise. The sound of percussion instruments occupies a wide frequency range but is narrow in the time domain, while the harmony (melody, chords, bass) is the opposite. In this system, it is necessary to focus on analyzing the harmony and melody information, so the noise of percussion instruments should be removed. In addition, musical instruments not only emit pure sounds, but also a series of higher-frequency harmonics and lower-frequency subharmonics, usually called overtones. This sound can easily confuse the original harmony, so it needs to be removed.

[0080] Tuning: In 2003, Sheh and Ellis discovered that some pop music tracks were not tuned to the standard A4=440HZ. In order to compensate for this problem in music recordings, this step tunes the audio to the standard tone.

[0081] Pitch extraction: This process matches the spectrogram to the human auditory system, that is, obtains the fixed note name corresponding to the frequency, and converts the spectrogram into a chromatogram. In addition, in order to facilitate the analysis of chords and melodies, the loudness factor is added to obtain a chromatogram based on loudness.

[0082] Octave merging and normalization: The octave change of notes in music has no effect on the basic properties of harmony, nor will it affect the mode. Therefore, when analyzing chords or modes, octaves need to be merged to reduce unnecessary effects.

[0083] Step 103: inputting the chromaticity features of the music audio to be identified into a mode recognition model to obtain the mode of the music audio to be identified; the mode recognition model is obtained by training a machine learning model using a training set.

[0084] The method for determining the mode recognition model mainly includes: preprocessing, chroma feature extraction and model building and training, such as Figure 2 The specific determination method is as follows.

[0085] (1) Obtain a training set; the training set includes training music audio and corresponding annotated modes.

[0086] (2) Extracting the chromaticity features of the training music audio.

[0087] Before extracting the chroma features of the training music audio, it is necessary to perform data enhancement and preprocessing operations on the input training music audio. The preprocessing operations include pre-emphasis processing, framing, windowing, and short-time Fourier transform operations. Figure 3 As shown. Digital music is usually sampled at a rate of 44,100 samples per second (CD quality). In the original form of this music audio, it cannot directly provide the frequency and harmonic information of the audio. There is evidence that the human auditory system performs a transformation from the time domain to the frequency domain, enabling people to perceive melodic and harmonic information. Similarly, the signal is converted into a representation that more directly reflects the frequency content. The Fourier transform only reflects the characteristics of the signal in the frequency domain and cannot analyze the signal in the time domain. In order to link the time domain and the frequency domain, the short-time Fourier transform (STFT) of the audio is usually used to calculate the frequency amplitude in the sliding window of the signal. It combines the characteristics of the spectrogram and the time domain waveform, and clearly shows the changes of the music spectrum over time, or it can be said to be a dynamic spectrum.

[0088] Therefore, in step (2), the training music audio is first subjected to data enhancement processing to obtain enhanced training music audio; the data enhancement processing includes speed conversion and pitch conversion. Then, the enhanced training music audio is subjected to preprocessing operations to obtain a spectrogram of the training music audio.

[0089] Data enhancement processing: Since the training model requires a huge amount of data, it is necessary to enhance the existing data to improve the versatility of the model. Since the mode is only related to the relative pitch of the music melody, and has nothing to do with the speed and absolute pitch of the music, the speed and pitch of the original music data are changed to expand the data volume. The training music audio is subjected to data enhancement processing to obtain enhanced music audio; the data enhancement processing includes speed change and pitch change, and the specific data enhancement processing process is as follows:

[0090] The speed of the training music audio is transformed according to 1.1, 1.2, 0.9 and 0.8 times the speed respectively to obtain speed-changed music audio; the speed-changed music audio includes the music audio before the speed change and the music audio after the speed change.

[0091] The pitch of the speed-changed music audio is raised and lowered by 1-3 semitones respectively to obtain the pitch-changed music audio; the pitch-changed music audio includes the music audio before the pitch-changed music audio and the music audio after the pitch-changed music audio. The pitch-changed music audio is determined as the enhanced training music audio.

[0092] The process of preprocessing the enhanced training music audio is similar to the preprocessing of the music audio to be identified in step 102, and will not be repeated here.

[0093] After obtaining the spectrum of the training music audio, the spectrum of the training music audio is converted into a chromatogram, and the chromatogram of the training music audio is used as the chromaticity feature of the training music audio. The conversion process of the spectrum to the chromatogram is as follows: Figure 4 As shown, the conversion process is similar to the conversion process in step 102 and will not be repeated here.

[0094] (3) The chromaticity features of the training music audio and the corresponding annotated modes are input into a machine learning model for training, and the trained machine learning model is determined as a mode recognition model.

[0095] Specifically: The concept of Chinese traditional music mode is not exactly the same as the concept of Western mode. It is difficult to design a completely standard set of rules to identify the mode. In addition, the same piece of music has different performance versions, which lead to inconsistent modes. Therefore, this embodiment uses a machine learning model to solve the problem of mode recognition. The chromatographic features and annotated modes of the music audio are input into the machine learning model to allow the model to learn the characteristics of different modes. After the model is trained, unknown music is input, and the chromatic features are extracted and then input into the model to finally obtain the predicted mode.

[0096] The modal characteristics of traditional music are closely related to the relative order of the chords and melodies of the entire song, and are highly time-related. Therefore, the Long-Short Time Memory (LSTM) model in the Recurrent Neural Network (RNN) model is used to deal with this problem.

[0097] RNN is a time-recursive network. This type of network is inspired by the cyclic connection of neurons in the human brain and stores information through a function structure similar to a cycle. Its cyclic structure is often used to handle tasks related to time series. Figure 5 As shown in the figure, Xt represents the input at time t, and ht represents the output at time t. Two different moments are connected by state information Ct, that is, the network has two inputs and one output at each moment. This simple cyclic time recursive structure will cause the problem of network gradient explosion as the number of layers increases, resulting in the network being equivalent to only the last few layers being useful.

[0098] LSTM is also a RNN model. RNN can only maintain short-term memory due to gradient vanishing. LSTM solves the gradient vanishing problem to a certain extent by introducing memory units and gate control units (combining short-term memory with long-term memory), and can learn long-term dependent information. LSTM adds forget gates, input gates, and output gates to the above RNN structure.

[0099] The forget gate controls how much of the information in the previous memory cell is forgotten:

[0100] f t =σ(W f ·[h t-1 , x t ]+b f );

[0101] Among them, f t represents the forget gate, f t is a vector, f tEach element in is in the range of [0, 1]; σ is the activation function, σ is a value between [0, 1]; W f is the weight corresponding to the forget gate; b f is the bias corresponding to the forget gate; x t Indicates the input event information at time t; h t-1 Represents the output value at time t-1.

[0102] The input gate controls how much of the current state to update:

[0103] i t =σ(W i ·[h t-1 , x t ]+b i );

[0104] Among them, i t represents the input gate; W i represents the weight corresponding to the input gate; b i is the bias corresponding to the input gate.

[0105] The output gate controls how much of the current memory unit is output:

[0106] o t =σ(W o [h t-1 , x t ]+b o );

[0107] Among them, t represents the output gate; W o represents the weight corresponding to the output gate; b o is the bias corresponding to the output gate.

[0108] LSTM controls the transmission state through the gate state, remembers the information that needs to be remembered for a long time, and forgets the unimportant information. It is suitable for many tasks that require long-term memory.

[0109] In the present invention, the structures of the two LSTMs are the same.

[0110] The Dropout layer is a structure that can be used to reduce overfitting of neural networks. Large neural networks are prone to overfitting. The dropout layer can randomly remove some neurons. The probability set in this system is 20%.

[0111] The machine learning model in this embodiment includes a first long short-term memory model layer (LSTM), a Dropout layer, a second long short-term memory model layer (LSTM), a full connection layer (Full Connection, FC) and a softmax function connected in sequence, such as Figure 6The input of the entire model is the chromaticity features of all songs and the marked modes.

[0112] In this embodiment, a total of 100 traditional pentatonic music are selected, including 48 Gong mode, 30 Yu mode, and 22 Zhi mode. The ratio of training set to test set is set to 8:2, that is, 80 songs are used for training and 20 songs are used for testing. After data enhancement, the total data is expanded to 500 songs, 400 of which are used as training sets and 100 are used as test sets. After the model training is completed, the test set is input, and the predicted labels are compared with the known labels. The result accuracy is 59.0%.

[0113] In practical applications, an implementation process of the five-tone music mode recognition method in the above embodiment is as follows: perform data enhancement and preprocessing operations on the marked music, and use data enhancement to increase the data volume; extract the chroma features of each song as the input of the subsequent model, take 80% as the training set, and 20% as the test set; use the song chroma features obtained in the previous step and the marked mode information as the model input training model, the model is: a layer of LSTM, a layer of Dropout layer, another layer of LSTM, and the last layer of fully connected layer, train the parameters of the model; input the test set into the trained model, obtain the test results, input the unmarked music, and obtain the mode recognized by the system. This method solves the problem of lack of a complete and standard database in the field of five-tone therapy, and can improve the accuracy and reliability of recognition.

[0114] The present invention also provides a five-tone music mode recognition system, comprising:

[0115] An audio acquisition module, used to acquire the music audio to be recognized;

[0116] A chromaticity feature extraction module, used to extract the chromaticity features of the music audio to be identified;

[0117] Mode recognition model module;

[0118] The mode recognition model training and recognition module is used to input the chromaticity features of the music audio to be recognized into the mode recognition model to obtain the mode of the music audio to be recognized; the mode recognition model is obtained by training the machine learning model using a training set.

[0119] The above four modules correspond to the operations of the four steps of the five-tone music mode recognition method of the present invention.

[0120] The above embodiments are only used to help understand the method and core idea of ​​the present invention; at the same time, for those skilled in the art, any changes made to the implementation methods according to the idea of ​​the present invention are within the scope of the present invention. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for identifying pentatonic music modes, characterized in that: The specific steps are: (1) Obtain the music audio to be identified; (2) extracting the chromaticity features of the music audio to be identified; (3) Build and train a mode recognition model; (4) Inputting the chromaticity features of the music audio to be identified into the trained mode recognition model, thereby obtaining the mode of the music audio to be identified; The mode recognition model described in step (3) is constructed, and the mode recognition model specifically includes a first long short-term memory model layer, a dropout layer, a second long short-term memory model layer, a fully connected layer and a softmax function connected in sequence; The training of the mode recognition model in step (3) specifically includes: (3.1) Obtaining a training set; the training set includes training music audio and corresponding annotated modes; (3.2) extracting chromaticity features of the training music audio; (3.3) Inputting the chromaticity features of the training music audio and the corresponding annotated modes into a machine learning model for training, and determining the trained machine learning model as a mode recognition model; The step (3.2) of extracting the chromaticity features of the training music audio specifically includes: (3.2.1) performing data enhancement processing on the training music audio to obtain enhanced training music audio; the data enhancement processing includes speed conversion and pitch conversion; (3.2.2) performing a preprocessing operation on the enhanced training music audio to obtain a frequency spectrum of the training music audio; (3.2.3) converting the spectrum of the training music audio into a chromatogram, wherein the chromatogram of the training music audio is a chromaticity feature of the training music audio; The data enhancement processing of the training music audio described in step (3.2.1) specifically includes: (3.2.1.1) The speed of the training music audio is changed according to 1.1, 1.2, 0.9 and 0.8 times the speed respectively, to obtain speed-changed music audio; the speed-changed music audio includes the music audio before the speed change and the music audio after the speed change; (3.2.1.2) Raising and lowering the pitch of the speed-changed music audio by 1-3 semitones respectively to obtain pitch-changed music audio; the pitch-changed music audio includes the music audio before pitch change and the music audio after pitch change; (3.2.1.3) Determine the pitch-changed music audio as enhanced training music audio.

2. The five-tone music mode recognition method according to claim 1, characterized in that: The step (2) of extracting the chromaticity features of the music audio to be identified specifically includes: (2.1) performing a preprocessing operation on the music audio to be identified to obtain a frequency spectrum of the music audio to be identified; (2.2) Converting the spectrum of the music audio to be identified into a chromatogram, wherein the chromatogram of the music audio to be identified is a chromaticity feature of the music audio to be identified.

3. The five-tone music mode recognition method according to claim 2, characterized in that: The pre-processing operation of the music audio to be recognized described in step (2.1) specifically includes: (2.1.1) performing pre-emphasis processing on the music audio to be identified to obtain emphasized music audio; (2.1.2) performing frame processing on the emphasized music audio to obtain framed music audio; (2.1.3) performing windowing processing on the framed music audio to obtain windowed music audio; (2.1.4) Performing a short-time Fourier transform on the windowed music audio to obtain a spectrum diagram of the music audio to be identified.

4. The five-tone music mode recognition method according to claim 3, characterized in that: The music audio after the emphasis in step (2.1.2) is subjected to frame processing, specifically including: The pre-emphasized music audio is divided into frames with 512 sampling points as one frame to obtain the divided music audio; the time length of each frame in the divided music audio is 11.61ms, and the overlap rate between frames is 50%.

5. The five-tone music mode recognition method according to claim 2, characterized in that: The step (2.2) of converting the spectrum of the music audio to be identified into a chromatogram specifically includes: (2.2.1) removing overtones from the spectrum of the music audio to be identified, to obtain a spectrum after removal; (2.2.2) tuning the removed spectrum to obtain a tuned spectrum; (2.2.3) converting the tuned spectrogram into an initial chromatogram using a loudness-based method; (2.2.4) The octaves of the initial chromatogram are merged, and the merged chromatogram is normalized to obtain the chromatogram of the music audio to be identified.

6. A recognition system based on the five-tone music mode recognition method according to claim 1, comprising: An audio acquisition module, used to acquire the music audio to be recognized; A chromaticity feature extraction module, used to extract the chromaticity features of the music audio to be identified; Mode recognition model module; A mode recognition model training and recognition module, used for inputting the chromaticity features of the music audio to be recognized into the mode recognition model to obtain the mode of the music audio to be recognized; the mode recognition model is obtained by training the machine learning model using the training set; The above four modules correspond to the operations of the four steps of the five-tone music mode recognition method.

Images