Music generation dance posture training method based on AI

Music features are extracted through AI technology and the Diffusion model is used to generate matching dance postures. Combined with machine learning optimization and real-time feedback, the problem of low dance choreography efficiency in the existing technology is solved, and dance performance that is more closely matched with music is achieved.

CN119049504AInactive Publication Date: 2024-11-29SHANGHAI KINSENA ARTIFICIAL INTELLIGENCE TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411321585.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-20
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing dance posture training methods lack close matching with musical style, rhythm and melody, resulting in long and inefficient dance choreography.

Method used

Using AI-based music-generated dance posture training method, the harmony, melody, beat and style characteristics of music are extracted through deep learning technology, combined with the Diffusion model to generate dance posture sequences that match the music, and using machine learning algorithms for optimization and real-time feedback Adjustment.

Benefits of technology

It improves the matching degree between dance and music, shortens the time of dance choreography, improves creative efficiency, and makes dance movements more coordinated with music style, rhythm and melody.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119049504A_ABST
    Figure CN119049504A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of computers, and discloses an AI-based music generation dance posture training method, which comprises the following steps of: 1, selecting and uploading a music file; step 2, music feature extraction; step 3, generation of a dance posture; step 4, dance posture optimization; 5, performing real-time feedback and simulation training; and 6, performing actual performance and improvement. According to the system, the deep learning technology is used for comprehensively analyzing music through the system, multiple levels of features such as harmony, melody, rhythm and style are extracted, the actual performance of a dancer is analyzed through a machine learning algorithm, real-time feedback is provided, the dancer is helped to continuously correct and optimize the dance posture in the training process through real-time adjustment, and the training effect of the dancer is improved. Therefore, the coordination of the dance and the music is improved, and the beneficial effects of improving the matching degree of the dance and the music and enabling the dance movement to be more coordinated with the music style, rhythm and melody are achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an AI-based music-generated dance posture training method. Background Art

[0002] With the continuous development of music processing technology and motion capture technology, music-based dance movement synthesis technology has gradually become a research hotspot in the field of music understanding and dance synthesis.

[0003] For example, a new method, device and storage medium for generating dance postures from music based on multimodal feature fusion, with publication number CN 117316129 A, belongs to the field of audio-driven dance postures. The network framework and method include two main stages: 1) The training stage decomposes the dance into a series of basic dance units, through which the model learns how to move. 2) The generation stage, the model learns how to create dance by seamlessly organizing multiple basic dance moves based on the input music. This method solves the challenging problem of converting music into dance by establishing a correspondence between music and dance, taking into account multiple factors such as the style and beat of music and dance. The model creates dance by learning basic dance units and seamlessly organizing dance moves. The generated results are authentic, diverse, consistent in style and rhythmic. This method provides new possibilities for dance creation and performance and has the potential to be applied in practice.

[0004] However, existing dance posture training methods mostly rely on manual choreography and the dancer's own experience, and lack a close match with the music style, rhythm and melody. Summary of the invention

[0005] Technical issues solved:

[0006] In view of the shortcomings of the prior art, the present invention provides an AI-based music-generated dance posture training method, which has the advantages of improving the matching degree between dance and music, making dance movements more coordinated with music style, rhythm and melody; shortening the time of dance choreography and improving the efficiency of dance creation, thereby solving the above-mentioned technical problems.

[0007] Technical solution:

[0008] To achieve the above object, the present invention provides the following technical solution: a method for training dance postures generated by music based on AI, comprising the following steps:

[0009] Step 1. Select and upload music files: Select or upload the music files to be matched. The music file format and quality must meet the system requirements. The requirements are: audio file format: WAV, MP3, FLAC or AAC; audio file sampling rate: 44.1kHz; audio file bit depth: 16-bit or 24-bit;

[0010] Step 2: Extract music features: Use deep learning technology to comprehensively analyze the music and extract the structural features of the music: harmony, melody, beat features and style features. These features are integrated to form a comprehensive feature representation of the music. The beat features include rhythm and beat, and the style features include music type and style.

[0011] Step 3: Dance posture generation: Input the extracted music features into the trained dance posture generation model, select the Diffusion model, and generate a dance posture sequence that matches the music style, rhythm, and melody based on the music features and the seed dance action sequence;

[0012] Step 4: Dance posture optimization: Optimize and adjust the generated dance posture sequence, use machine learning algorithms to refine the movements, and perform multiple rounds of corrections to optimize the final result;

[0013] Step 5: Real-time feedback and simulation training: When dancers are doing simulation training, sensors are used to monitor the dancers’ movements in real time, and machine learning algorithms are used to analyze the dancers’ actual performances, providing real-time feedback and suggestions to help dancers adjust and optimize their dance postures.

[0014] Step 6: Actual performance and improvement: Dancers perform according to the optimized dance posture sequence, collect performance data, evaluate the matching degree between dance movements and music, and further adjust the dance posture sequence based on feedback.

[0015] Preferably, the audio file format requirement expression in step 1 is:

[0016] Format∈WAV,MP3,FLAC,AAC

[0017] The audio sampling rate requirement expression is: sampling rate = 44.1kHz;

[0018] The audio bit depth requirement expression is:

[0019] Bit depth ∈ 16 bit, 24 bit.

[0020] Preferably, the harmony feature extraction expression in step 2 is:

[0021] H=extract harmony(x)

[0022] Among them, x is the audio signal, H is the harmony feature, extract harmony(x) is a function that receives an audio signal x and returns the harmonic features;

[0023] The melody feature extraction expression is:

[0024] M=extract melody(x)

[0025] Among them, M is the melody feature, extract melody(x) is a function that receives an audio signal x and returns the melody features;

[0026] The beat feature extraction expression is:

[0027] B = rhythm, beat

[0028] Explanation: B represents the beat feature extracted from the audio, including two sub-features: rhythm and beat;

[0029] The style feature extraction expression is:

[0030] S = Music type, style

[0031] Explanation: S represents the style feature extracted from the audio, including two sub-features: music type and style. Preferably, the comprehensive feature fusion expression in step 2 is:

[0032] F = comprehensive characteristics (H, M, B, S)

[0033] Explanation: F represents the comprehensive music feature representation obtained by integrating the harmony feature H, melody feature M, beat feature B and style feature S.

[0034] Preferably, the expression of inputting music features into the dance posture generation model in step 3 is:

[0035] D=G(M,B,S,S d )

[0036] Where: D represents the generated dance posture sequence; G is the dance posture generation model, M is the melody feature, B is the beat feature including rhythm and beat, S is the style feature, S d is a sequence of seed dance moves; model G generates a sequence of dance poses D that matches the music features based on these inputs.

[0037] Preferably, in step 3, the music features are input into the Diffusion model:

[0038] D = DiffusionModel(M,B,S,S d )

[0039] Where: D is the generated dance posture sequence; DiffusionModel is the Diffusion model used to generate dance postures; M is the melody feature vector, which represents the melody information of the music; B is the beat feature vector, including rhythm and beat; S is the style feature vector, which represents the style of the music; S dis a seed dance action sequence, used as the initial reference for generating dance;

[0040] The specific application expression of the Diffusion model is:

[0041] Initial state:

[0042] x T =S d

[0043] Where: x T Represents the initial state of the Diffusion model, that is, the seed dance action sequence S d ;

[0044] Noise addition and denoising process:

[0045] x t =xt +1 ―αt(xt +1 ―NoisedPose(x t+1 ,M,B,S))

[0046] Where: x t represents the dance posture at time step t, x t+1 is the dance posture at time step t+1; α t is the attenuation coefficient of the time step, NoisedPose(x t+1 ,M,B,S) represents the noise adding function based on music features.

[0047] Preferably, the Diffusion model in step 3 further includes an optimization target, and the optimization target expression is:

[0048] L=|x0―PoseMatch(D,M,B,S)| 2

[0049] Where: L represents the optimization objective function, which is used to minimize the difference between the generated dance pose sequence D and the target dance pose adjusted according to the music features; PoseMatch(D,M,B,S) is the function for matching the target dance pose.

[0050] Preferably, the dance posture sequence expression generated in step 4 is:

[0051] D raw =DiffusionModel(M,B,S,S d )

[0052] Where: D raw is the initially generated dance posture sequence; DiffusionModel is the Diffusion model used to generate the dance posture sequence;

[0053] The optimization target expression in step 4 is:

[0054]

[0055] Where: L is the overall optimization objective function; x i is the dance posture of the i-th frame; PoseMatch(x i ,M,B,S) is the target posture matching function based on music features; λ is the weight parameter for balancing posture matching and motion smoothing; |x i+1 ―x i | 2 It is a constraint term used for smooth motion and ensures the continuity of the dance posture sequence.

[0056] Preferably, the real-time action monitoring expression in step 5 is:

[0057] x sno (t) = SensorData(t)

[0058] Where: x sno (t): the actual movement data of the dancer obtained from the sensor at time t; SensorData(t): represents the real-time data function collected from the sensor.

[0059] Preferably, the actual performance data collection expression in step 6 is:

[0060] x promne(t) =PerformanceData(t)

[0061] Where: x promne(t) is the actual performance action data collected at time t; PerformanceData(t) represents the data function obtained from the actual performance;

[0062] The matching evaluation expression is:

[0063] e promne(t) =x promne(t) ―x tre (t)

[0064] Where: e promne(t) is the action error at time t, that is, the difference between the actual performance action and the target action; x promne(t) is the optimized target dance posture sequence.

[0065] Compared with the prior art, the present invention provides an AI-based music generation dance posture training method, which has the following beneficial effects:

[0066] 1. The present invention uses deep learning technology to comprehensively analyze music through the system, and extracts multiple levels of features such as harmony, melody, beat and style. The features include the structural information, rhythm and beat of the music, as well as the type and style of the music. This meticulous feature extraction ensures a comprehensive understanding of the music, so that the generated dance postures can more accurately reflect the various characteristics of the music. In step three, the extracted music features are input into the trained Diffusion model, which generates a dance posture sequence that matches the music style and rhythm by combining the melody, beat and style features of the music and the seed dance movement sequence. This process ensures that the generated dance movements can match all aspects of the music. In step 4, the generated dance posture sequence is optimized and adjusted, and the machine learning algorithm is used to refine the movements to ensure that the generated dance is more in line with the characteristics of the music. This optimization process continuously improves the matching degree between dance movements and music through multiple rounds of corrections, making the dance performance more natural and smooth. In step 5, the system uses sensors to monitor the dancer's movements in real time, analyzes the dancer's actual performance through machine learning algorithms, and provides real-time feedback. This real-time adjustment helps the dancer to continuously correct and optimize the dance posture during training, thereby improving the coordination between dance and music, achieving the beneficial effect of improving the matching degree between dance and music and making the dance movements more coordinated with the music style, rhythm and melody.

[0067] 2. The present invention uses a system to automatically extract the harmony, melody, beat and style features of music through deep learning technology, and inputs these features into the Diffusion model to generate a dance posture sequence. This process replaces the traditional manual choreography steps, making dance choreography more automated and efficient. Application of the Diffusion model: The Diffusion model can consider the various features of music when generating a dance posture sequence, and automatically adjust the generated dance movements to match the style and rhythm of the music. This reduces the manual adjustment and creation time in dance choreography. The generated dance posture sequence is optimized by a machine learning algorithm, and the movements are refined and corrected for multiple rounds. The automated optimization process greatly reduces the time for manual adjustment and improves the efficiency of dance choreography. In the simulation training stage, the system uses sensors to monitor the dancer's movements in real time and provide real-time feedback. The instant adjustment helps the dancer quickly improve the dance movements, shortens the training time, and improves the creation efficiency, achieving the beneficial effect of shortening the time for dance choreography and improving the efficiency of dance creation. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of the method of the present invention; DETAILED DESCRIPTION

[0069] 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.

[0070] See also Figure 1 , a music generation dance posture training method based on AI, comprising the following steps:

[0071] Step 1. Select and upload music files: Select or upload the music files to be matched. The music file format and quality must meet the system requirements. The requirements are: audio file format: WAV, MP3, FLAC or AAC; audio file sampling rate: 44.1kHz; audio file bit depth: 16-bit or 24-bit;

[0072] Step 2: Extract music features: Use deep learning technology to comprehensively analyze the music and extract the structural features of the music: harmony, melody, beat features and style features. These features are integrated to form a comprehensive feature representation of the music. The beat features include rhythm and beat, and the style features include music type and style.

[0073] Step 3: Dance posture generation: Input the extracted music features into the trained dance posture generation model, select the Diffusion model, and generate a dance posture sequence that matches the music style, rhythm, and melody based on the music features and the seed dance action sequence;

[0074] Step 4: Dance posture optimization: Optimize and adjust the generated dance posture sequence, use machine learning algorithms to refine the movements, and perform multiple rounds of corrections to optimize the final result;

[0075] Step 5: Real-time feedback and simulation training: When dancers are doing simulation training, sensors are used to monitor the dancers’ movements in real time, and machine learning algorithms are used to analyze the dancers’ actual performances, providing real-time feedback and suggestions to help dancers adjust and optimize their dance postures.

[0076] Step 6: Actual performance and improvement: Dancers perform according to the optimized dance posture sequence, collect performance data, evaluate the matching degree between dance movements and music, and further adjust the dance posture sequence based on feedback.

[0077] In step two, the system uses deep learning technology to comprehensively analyze the music and extracts features at multiple levels, such as harmony, melody, beat, and style. The features include the structural information, rhythm and beat of the music, as well as the type and style of the music. This detailed feature extraction ensures a comprehensive understanding of the music, so that the generated dance postures can more accurately reflect the various characteristics of the music.

[0078] Dance pose generation and music matching: In step three, the extracted music features are input into the trained Diffusion model, which generates a dance pose sequence that matches the music style and rhythm by combining the melody, beat and style features of the music with the seed dance move sequence. This process ensures that the generated dance moves can be coordinated with all aspects of the music.

[0079] Optimization and adjustment: In step 4, the generated dance posture sequence is optimized and adjusted, and the movements are refined using machine learning algorithms to ensure that the generated dance is more in line with the characteristics of the music. This optimization process continuously improves the match between dance movements and music through multiple rounds of corrections, making the dance performance more natural and smooth.

[0080] Real-time feedback and simulated training: In step five, the system uses sensors to monitor the dancer's movements in real time, analyzes the dancer's actual performance through machine learning algorithms, and provides real-time feedback. This real-time adjustment helps the dancer to continuously correct and optimize the dance posture during training, thereby improving the coordination between dance and music.

[0081] Actual performance and improvement: In step 6, the dancers actually perform according to the optimized dance posture sequence. The system collects performance data and evaluates the matching degree between the dance movements and the music. The dance posture sequence is further adjusted based on the evaluation results. This feedback mechanism ensures that the dance movements are highly coordinated with the music.

[0082] Specifically, the audio file format requirement expression in step 1 is:

[0083] Format∈WAV,MP3,FLAC,AAC

[0084] The audio sampling rate requirement expression is: sampling rate = 44.1kHz;

[0085] The audio bit depth requirement expression is:

[0086] Bit depth ∈ 16 bit, 24 bit.

[0087] Specifically, the harmony feature extraction expression in step 2 is:

[0088] H=extract harmony(x)

[0089] Among them, x is the audio signal, H is the harmony feature, extract harmony(x) is a function that receives an audio signal x and returns the harmonic features;

[0090] The melody feature extraction expression is:

[0091] M=extract melody(x)

[0092] Among them, M is the melody feature, extract melody(x) is a function that receives an audio signal x and returns the melody features;

[0093] The beat feature extraction expression is:

[0094] B = rhythm, beat

[0095] Explanation: B represents the beat feature extracted from the audio, including two sub-features: rhythm and beat;

[0096] The style feature extraction expression is:

[0097] S = Music type, style

[0098] Explanation: S represents the style feature extracted from the audio, including two sub-features: music type and style. Specifically, the comprehensive feature fusion expression in step 2 is:

[0099] F = comprehensive characteristics (H, M, B, S)

[0100] Explanation: F represents the comprehensive music feature representation obtained by integrating the harmony feature H, melody feature M, beat feature B and style feature S.

[0101] Specifically, the expression of inputting music features into the dance posture generation model in step 3 is:

[0102] D=G(M,B,S,S d )

[0103] Where: D represents the generated dance posture sequence; G is the dance posture generation model, M is the melody feature, B is the beat feature including rhythm and beat, S is the style feature, S d is a sequence of seed dance moves; model G generates a sequence of dance poses D that matches the music features based on these inputs.

[0104] Specifically, the music features in step 3 are input into the Diffusion model:

[0105] D = DiffusionModel(M,B,S,S d )

[0106] Where: D is the generated dance posture sequence; DiffusionModel is the Diffusion model used to generate dance postures; M is the melody feature vector, which represents the melody information of the music; B is the beat feature vector, including rhythm and beat; S is the style feature vector, which represents the style of the music; S d is a seed dance action sequence, used as the initial reference for generating dance;

[0107] The specific application expression of the Diffusion model is:

[0108] Initial state:

[0109] x T =S d

[0110] Where: x T Represents the initial state of the Diffusion model, that is, the seed dance action sequence S d ;

[0111] Noise addition and denoising process:

[0112] x t =x t+1 ―α t (x t+1 ―NoisedPose(x t+1 ,M,B,S))

[0113] Where: x t represents the dance posture at time step t, x t+1 is the dance posture at time step t+1; α t is the attenuation coefficient of the time step, NoisedPose(x t+1 ,M,B,S) represents the noise adding function based on music features.

[0114] Specifically, the Diffusion model in step 3 also includes an optimization target, and the optimization target expression is:

[0115] L=|x0―PoseMatch(D,M,B,S)| 2

[0116] Where: L represents the optimization objective function, which is used to minimize the difference between the generated dance pose sequence D and the target dance pose adjusted according to the music features; PoseMatch(D,M,B,S) is the function for matching the target dance pose.

[0117] Specifically, the dance posture sequence expression generated in step 4 is:

[0118] D raw=DiffusionModel(M,B,S,S d )

[0119] Where: D raw is the initially generated dance posture sequence; DiffusionModel is the Diffusion model used to generate the dance posture sequence;

[0120] The optimization target expression in step 4 is:

[0121]

[0122] Where: L is the overall optimization objective function; x i is the dance posture of the i-th frame; PoseMatch(x i ,M,B,S) is the target posture matching function based on music features; λ is the weight parameter for balancing posture matching and motion smoothing; |x i+1 ―x i | 2 It is a constraint term used for smooth motion and ensures the continuity of the dance posture sequence.

[0123] Specifically, the real-time action monitoring expression in step 5 is:

[0124] x sno (t) = SensorData(t)

[0125] Where: x sno (t): the actual movement data of the dancer obtained from the sensor at time t; SensorData(t): represents the real-time data function collected from the sensor.

[0126] Specifically, the actual performance data collection expression in step 6 is:

[0127] x promne(t) =PerformanceData(t)

[0128] Where: x promne(t) is the actual performance action data collected at time t; PerformanceData(t) represents the data function obtained from the actual performance;

[0129] The matching evaluation expression is:

[0130] e promne(t) =x promne(t) ―x tre (t)

[0131] Where: e promne(t) is the action error at time t, that is, the difference between the actual performance action and the target action; xpromne(t) is the optimized target dance posture sequence.

[0132] The system automatically extracts the harmony, melody, beat and style features of the music through deep learning technology, and inputs these features into the Diffusion model to generate a dance posture sequence. This process replaces the traditional manual choreography steps, making dance choreography more automated and efficient.

[0133] Application of Diffusion Model: Diffusion model can consider various characteristics of music when generating dance posture sequences, and automatically adjust the generated dance movements to match the style and rhythm of the music. This reduces the manual adjustment and creation time in dance choreography.

[0134] Machine learning optimization: The generated dance posture sequence is optimized through machine learning algorithms, which refines the movements and performs multiple rounds of corrections. This automated optimization process greatly reduces the time for manual adjustments and improves the efficiency of dance choreography.

[0135] Real-time feedback system: During the simulation training stage, the system uses sensors to monitor the dancers' movements in real time and provide real-time feedback. This instant adjustment helps dancers quickly improve their dance movements, shortens training time, and improves creative efficiency.

[0136] Comprehensive feature fusion: The system automatically generates dance postures that conform to the characteristics of the music by integrating the harmony, melody, beat and style characteristics of the music. This automated feature fusion reduces the dependence on dance choreographers and simplifies the creative process.

[0137] Automatic evaluation and improvement: Through real-time monitoring and performance data collection, the system can automatically evaluate the matching degree between dance movements and music, and further adjust dance postures based on feedback. This automatic evaluation mechanism reduces the need for manual evaluation and improves creative efficiency.

[0138] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A music-generated dance posture training method based on AI, characterized in that: The following steps are involved: Step 1. Select and upload music files: Select or upload the music files to be matched. The music file format and quality must meet the system requirements. The requirements are: audio file format: WAV, MP3, FLAC or AAC; audio file sampling rate: 44.1kHz; audio file bit depth: 16-bit or 24-bit; Step 2: Extract music features: Use deep learning technology to comprehensively analyze the music and extract the structural features of the music: harmony, melody, beat features and style features. These features are integrated to form a comprehensive feature representation of the music. The beat features include rhythm and beat, and the style features include music type and style. Step 3: Dance posture generation: Input the extracted music features into the trained dance posture generation model, select the Diffusion model, and generate a dance posture sequence that matches the music style, rhythm, and melody based on the music features and the seed dance action sequence; Step 4: Dance posture optimization: Optimize and adjust the generated dance posture sequence, use machine learning algorithms to refine the movements, and perform multiple rounds of corrections to optimize the final result; Step 5: Real-time feedback and simulation training: When dancers are doing simulation training, sensors are used to monitor the dancers’ movements in real time, and machine learning algorithms are used to analyze the dancers’ actual performances, providing real-time feedback and suggestions to help dancers adjust and optimize their dance postures. Step 6: Actual performance and improvement: Dancers perform according to the optimized dance posture sequence, collect performance data, evaluate the matching degree between dance movements and music, and further adjust the dance posture sequence based on feedback.

2. The method for training dance postures using music generated by AI according to claim 1, characterized in that: The audio file format requirement expression in step 1 is: Format∈WAV,MP3,FLAC,AAC The audio sampling rate requirement expression is: sampling rate = 44.1kHz; The audio bit depth requirement expression is: Bit depth ∈ 16 bit, 24 bit.

3. The AI-based music-generated dance posture training method according to claim 1, characterized in that: The harmony feature extraction expression in step 2 is: H=extract harmony(x) Among them, x is the audio signal, H is the harmony feature, extract harmony(x) is a function that receives an audio signal x and returns the harmonic features; The melody feature extraction expression is: M=extract melody(x) Among them, M is the melody feature, extract melody(x) is a function that receives an audio signal x and returns the melody features; The beat feature extraction expression is: B = rhythm, beat Explanation: B represents the beat feature extracted from the audio, including two sub-features: rhythm and beat; The style feature extraction expression is: S = Music type, style Explanation: S represents the style feature extracted from the audio, including two sub-features: music type and style.

4. The method for training dance postures based on music generation based on AI according to claim 3, characterized in that: The comprehensive feature fusion expression in step 2 is: F = comprehensive characteristics (H, M, B, S) Explanation: F represents the comprehensive music feature representation obtained by integrating the harmony feature H, melody feature M, beat feature B and style feature S.

5. The AI-based music-generated dance posture training method according to claim 1, characterized in that: The expression of inputting music features into the dance posture generation model in step 3 is: D=G(M,B,S,S d ) Where: D represents the generated dance posture sequence; G is the dance posture generation model, M is the melody feature, B is the beat feature including rhythm and beat, S is the style feature, S d is a sequence of seed dance moves; model G generates a sequence of dance poses D that matches the music features based on these inputs.

6. The AI-based music-generated dance posture training method according to claim 5, characterized in that: In step 3, the music features are input into the Diffusion model: D=DiffusionModel(M,B,S,S d ) Where: D is the generated dance posture sequence; DiffusionModel is the Diffusion model used to generate dance postures; M is the melody feature vector, which represents the melody information of the music; B is the beat feature vector, including rhythm and beat; S is the style feature vector, which represents the style of the music; S d is a seed dance action sequence, used as the initial reference for generating dance; The specific application expression of the Diffusion model is: Initial state: x T =S d Where: x T Represents the initial state of the Diffusion model, that is, the seed dance action sequence S d ; Noise addition and denoising process: x t =x t+1 ―α t (x t+1 ―NoisedPose(x t+1 ,M,B,S)) Where: x t represents the dance posture at time step t, x t+1 is the dance posture at time step t+1; α t is the attenuation coefficient of the time step, NoisedPose(x t+1 ,M,B,S) represents the noise adding function based on music features.

7. The AI-based music-generated dance posture training method according to claim 6, characterized in that: The Diffusion model in step 3 also includes an optimization target, and the optimization target expression is: L=|x0―PoseMatch(D,M,B,S)| 2 Where: L represents the optimization objective function, which is used to minimize the difference between the generated dance pose sequence D and the target dance pose adjusted according to the music features; PoseMatch(D,M,B,S) is the function for matching the target dance pose.

8. The AI-based music-generated dance posture training method according to claim 7, characterized in that: The dance posture sequence expression generated in step 4 is: D raw =DiffusionModel(M,B,S,S d ) Where: D raw is the initially generated dance posture sequence; DiffusionModel is the Diffusion model used to generate the dance posture sequence; The optimization target expression in step 4 is: Where: L is the overall optimization objective function; x i is the dance posture of the i-th frame; PoseMatch(x i ,M,B,S) is the target posture matching function based on music features; λ is the weight parameter for balancing posture matching and motion smoothing; |x i+1 ―x i | 2 It is a constraint term used for smooth motion and ensures the continuity of the dance posture sequence.

9. The AI-based music-generated dance posture training method according to claim 8, characterized in that: The real-time action monitoring expression in step 5 is: x sno (t)=SensorData(t) Where: x sno (t): the actual movement data of the dancer obtained from the sensor at time t; SensorData(t): represents the real-time data function collected from the sensor.

10. The AI-based music-generated dance posture training method according to claim 9, characterized in that: The actual performance data collection expression in step 6 is: x promne(t) =PerformanceData(t) Where: x promne(t) is the actual performance action data collected at time t; PerformanceData(t) represents the data function obtained from the actual performance; The matching evaluation expression is: e promne(t) =x promne(t) ―x tre (t) Where: e promne(t) is the action error at time t, that is, the difference between the actual performance action and the target action; x promne(t) is the optimized target dance posture sequence.