AST model-based music singing scoring system and application

Abstract

The invention relates to the technical field of singing scoring systems, and particularly discloses a stylus scoring system based on an AST model and application, and the system specifically comprises the following steps: 1, quantification of a stylus singing skill: dividing a singing work of a singer into ten fingers, and collecting a singing data set of the singer; 2, for the ten indexes, an international beautiful sound competition scoring method and a five-division system are adopted, 1 is the best, 2 is the better, 3 is the general, 4 is the poorer, and 5 is the worst; 3, after the data set is collected, a famous owner of the beauty industry scores the data set, and the scoring accuracy is ensured; according to the method, the spectrogram of the singer singing audio is analyzed through the artificial intelligence technology, comparison with the original music is not needed, and the application range is wider.

Description

Technical Field

[0001] The present invention relates to the technical field of singing scoring systems, in particular to a bel canto singing scoring system based on an AST model and its application. Background Art

[0002] Traditional singing scoring software is designed to compare the degree of fit between the user and the original song, and give points based on the degree of fit. Such scoring is extremely dependent on the original song, and the degree of freedom and adaptability are not high. In terms of musical aesthetics, it only uses the original song as a single standard, and the evaluation of vocal singing is not comprehensive enough. Most of the existing scoring systems are aimed at popular singing methods and techniques, and rarely involve bel canto singing methods. Summary of the invention

[0003] In view of the technical problems mentioned in the background technology that the scoring is extremely dependent on the original song and has low freedom and adaptability, the present invention provides a bel canto scoring system and application based on the AST model.

[0004] The technical solution adopted by the present invention is: a bel canto singing scoring system based on the AST model, which specifically includes the following steps:

[0005] Step 1: Quantification of Bel Canto singing skills: First, the singer's singing works are divided into ten fingers, and the singer's singing data set is collected;

[0006] Step 2: For these ten indicators, the international bel canto competition scoring method is used, with a five-point system, where 1 is the best, 2 is better, 3 is average, 4 is poor, and 5 is the worst;

[0007] Step 3: After the data set is collected, it is scored by famous experts in the bel canto industry to ensure the accuracy of the scoring;

[0008] Step 4: AI scoring and analysis of singing works: Use AI algorithms to build a model, convert the sound into a spectrum, and input it into the model for training;

[0009] Step 5: Using the AST model, through training of multiple data sets, the characteristics of each indicator data are extracted to achieve accurate scoring of bel canto singing works, and the scoring results are displayed using Kiviat graphs;

[0010] Step six, after scoring, use neural network as an analysis tool to analyze the changes in scores of the same singer singing the same piece at different times, analyze their correlation, compare the quality of their previous and subsequent performances, and quantitatively analyze their performance changes.

[0011] The present invention is further configured as follows: in step one, the singer's performance is divided into ten indicators, and the ten indicators are: Vibrato (tremolo), Throat (laryngeal position), Position (sound position), Open (degree of cavity opening), Clean (degree of purity of tone), Resonate (resonance), Unify (degree of sound unity), Falsetto (falsetto), Chest (chest voice), and Nasal (nasal sound).

[0012] The present invention is further configured to apply the bel canto scoring system based on the AST model to singing scoring software.

[0013] The beneficial effects of the present invention are: in the present invention, the spectrum of the singer's singing audio is analyzed by artificial intelligence technology, and there is no need to compare with the original song, so the scope of application is wider. Secondly, the scoring criteria in the present invention refer to the scoring criteria of the International Bel Canto Competition, and the singer's singing level is judged from the dimensions of ten technical indicators, which is more refined, objective and professional than the previous scoring criteria. Furthermore, the AST model used in this article has extremely high accuracy after training with a large number of high-quality training sets, and the average accuracy exceeds 90%. This lays a good foundation for scoring and comparing the differences in singing before and after the same singer. Finally, by quantitative means, the effects of singing before and after can be accurately and efficiently compared from a mathematical level, which can effectively assist vocal teachers in training and adjusting teaching plans, and promote the development of vocal education to a more advanced level. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Schematic diagram of the score distribution of each bel canto singing technique in the data set of the present invention;

[0015] Figure 2 This is a schematic diagram of the AST model result analysis in the present invention;

[0016] Figure 3 Schematic diagram of the accuracy and loss function of the AST model training in the present invention Figure 1 ;

[0017] Figure 4 This is a schematic diagram of the accuracy and loss function of the AST model test in the present invention. Figure 2 ;

[0018] Figure 5 It is a schematic diagram of the confusion matrix of the AST model of the present invention under various data set allocations when the score is 1-5;

[0019] Figure 6 It is a schematic diagram of the AST model in the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with 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.

[0021] In order to solve the problems existing in the background technology, the present application proposes the following technical solutions: 1. A bel canto scoring system based on the AST model, characterized in that it specifically includes the following steps:

[0022] Step 1: Quantification of Bel Canto singing skills: First, the singer's singing works are divided into ten fingers, and the singer's singing data set is collected;

[0023] Step 2: For these ten indicators, the international bel canto competition scoring method is used, with a five-point system, where 1 is the best, 2 is better, 3 is average, 4 is poor, and 5 is the worst;

[0024] Step 3: After the data set is collected, it is scored by famous experts in the bel canto industry to ensure the accuracy of the scoring;

[0025] Step 4: AI scoring and analysis of singing works: Use AI algorithms to build a model, convert the sound into a spectrum, and input it into the model for training;

[0026] Step 5: Using the AST model, through training of multiple data sets, the characteristics of each indicator data are extracted to achieve accurate scoring of bel canto singing works, and the scoring results are displayed using Kiviat graphs;

[0027] Step six, after scoring, use neural network as an analysis tool to analyze the changes in scores of the same singer singing the same piece at different times, analyze their correlation, compare the quality of their previous and subsequent performances, and quantitatively analyze their performance changes.

[0028] Among them, in step one, the singer's singing works are divided into ten indicators, and the ten indicators are: Vibrato, Throat, Position, Open, Clean, Resonate, Unify, Falsetto, Chest, and Nasal.

[0029] The above technical solution is explained as follows: the source of audio materials used for machine learning data set training:

[0030] 1. This dataset selects the audio materials of many soprano singers, which can be divided into three different levels:

[0031] 1. High-quality recordings of world-renowned top soprano singers (such as Edith Gruberova, Sandrine Piau, Diana Damrau, usw), as well as excellent recordings of professional singers from the world's top opera houses. Their singing is at the world's top level.

[0032] 2. Recordings of singing and voice training of undergraduate, master and doctoral students (soprano) from German and Austrian music academies and Chinese music colleges (including the University of Music and Performing Arts Vienna MDW, Hamburg University of Music, Berlin University of the Arts, Stuttgart Conservatory of Music, Hannover Conservatory of Music, Shanghai Conservatory of Music, China Conservatory of Music, etc.). Their singing is at a professional level.

[0033] 3. Recording materials for beginners of Bel Canto singing. Their singing is at the entry level.

[0034] 4. Recordings of untrained bel canto singers who did not use scientific singing methods.

[0035] 2. Dataset Scoring Criteria

[0036] The scoring criteria for this dataset were developed by the vocal performance professor team of the University of Music and Performing Arts Vienna (MDW) and vocal professors from many professional universities at home and abroad. From the perspective of bel canto singing, with reference to the scoring criteria for singers set by many international vocal competitions and the important technical characteristics of bel canto vocal teaching in universities, the following 10 technical elements for soprano voices are proposed:

[0037] 1. Vibrato:

[0038] vibrato:

[0039] Definition: The musical term vibrato is translated as "kneading fingers" in violin performance, and refers to the vibration of sound in vocal music. However, it is easily confused with other similar terms such as trill, tremolo, vibration (vibrazione), etc., so it is sometimes translated as "micro-tremor". The micro-tremor of normal human voice singing is generally six to seven times per second, which usually gives people a sense of stability, solemnity, tranquility and liveliness. A slightly faster micro-tremor may make people feel excited and excited, but too fast will make people feel uneasy. This kind of overly fast micro-tremor is called "small shaking", "broken shaking" or "tremolo" in vocal music. This is caused by the tension of the muscles in the front or back of the jaw or other muscles in the throat. It is usually caused by the singer's pursuit of an overly "bright" tone and ignoring the relaxation, roundness and smoothness of the voice. Too slow micro-tremor gives people a feeling of instability and inaccuracy, which is called "shaking" in vocal music. Generally speaking, this is caused by the tongue dropping too low and the tongue root pressing down, which is often related to the singer's excessive pursuit of a strong timbre or a loud volume. There is also a "straight" voice without tremor, which may be caused by weak vocal ability or stiff throat muscles from the perspective of vocalization skills. This voice gives people a feeling of incompetence or dullness. However, as a means of artistic expression, it can be used occasionally. For example, when expressing a scene of madness or shock, the singing voice at the beginning can be without tremor to adapt to the needs of the character, situation and atmosphere. Whether the tremor is normal, smooth and stable is often an effective means to identify the pros and cons of the vocalization method in vocal teaching. The tremor is also one of the important factors in maintaining the unity of sound and the balance and harmony of the chorus in a chorus. In a chorus, as long as the tremor of one person's voice is very different from that of others ("shaking" or "shaking"), this person will stand out. If this technical problem is not solved, it will destroy the chorus effect. The tremor of the human voice is one of the powerful means of expression in singing.

[0040] Scoring criteria:

[0041] 1 point: The vibrato is stable, harmonious, extremely natural and beautiful, and completely in line with the musical emotion and song background.

[0042] 2 and 3 points: The sound is natural and beautiful, but the vibration frequency is slightly too fast or too slow, or the amplitude is slightly too large. 4 points: The sound is unnatural, the vibration frequency is obviously too fast or too slow, or the vibration amplitude is too large, or even shaking. Or the sound is straightened, with very little vibration. 5 points: Excessive shaking or no vibration at all.

[0043] 2. Throat:

[0044] The position of the larynx plays an important role in vocal singing.

[0045] The correct position of the larynx can ensure the smooth phonation of the voice. The position of the larynx plays a key role in the control of airflow. It affects the tension and vibration frequency of the vocal cords, which directly affects the production of sound and the effect of phonation. When the larynx is in the proper position, the sound can be transmitted through the vocal cords more easily, keeping the timbre clear and bright. Secondly, the position of the larynx has an important influence on the timbre and quality of the voice. Different larynx positions will produce different resonance effects, thus affecting the timbre characteristics of the voice. By adjusting the position of the larynx, the singer can control the brightness, warmth and fullness of the voice, as well as meet the requirements of songs of different styles. In addition, the position of the larynx is also directly related to the volume and stability of the voice. The correct position of the larynx can help singers make full use of the airflow and make the voice generate enough energy, thereby achieving volume control and adjustment. At the same time, a stable larynx position can help singers maintain the stability and durability of the voice, and maintain the quality and expressiveness of the voice during long singing.

[0046] Scoring criteria: 1 point, correct laryngeal position, 2 points, slightly higher larynx position, 3 points, obviously higher or lower larynx position, 4 points, raised larynx or pressed larynx position, 5 points, high and squeezed larynx position

[0047] 3. Position

[0048] Sound location,

[0049] In Bel Canto, "voice position" refers to the position of the larynx and respiratory system that the singer adjusts to achieve the best sound production and vocalization effect. This concept covers several aspects: Laryngeal position adjustment: The singer needs to accurately control the height and opening degree of the larynx. The position of the larynx has an important impact on the production and quality of the sound. Generally speaking, the larynx is in a proper position to ensure smooth airflow through the vocal cords, making the sound more clear and bright. The degree of opening of the larynx also affects the size and resonance of the resonance cavity, thereby affecting the timbre and texture of the sound. Breathing control: Correct breathing is a crucial part of Bel Canto singing. Singers need to learn to control the depth, speed and stability of breathing to ensure sufficient and stable airflow when the sound is produced. Correct breathing can help singers achieve stable sound production and continuity, and help control the volume and intensity of the sound. Sound projection: Sound position also involves the projection and positioning of the sound. Singers need to learn to project the sound to the appropriate location, such as a specific area of ​​the head, nasal cavity or mouth, to achieve the best resonance effect and clear communication of the sound. With proper projection, the sound can be better perceived and appreciated by the audience, while also reducing tension and stress in the throat.

[0050] Scoring criteria: 1 point: Maintaining the correct voice position in all ranges and notes; 2 points: The correct position accounts for more than 90% of the singing; 3 points: The correct position accounts for more than 70% of the singing; 4 points: The correct position accounts for more than 60% of the singing; 5 points: The correct position accounts for less than 50% of the singing

[0051] 4. open

[0052] In Bel Canto, cavity opening refers to the state in which the singer adjusts the throat and respiratory system so that the sound can vibrate and resonate freely in the mouth and throat. This state enables the sound to fully utilize the resonance cavity, including the mouth, nasal cavity and throat, to produce a fuller, brighter and clearer timbre. Specifically, the process of cavity opening includes the following aspects: 1. Opening of the throat: The singer needs to ensure that the throat is in a proper open state so that the vocal cords can vibrate freely and it is also conducive to the smooth flow of air. The degree of opening of the throat should be moderate, neither too tight nor too loose, to ensure the stability and quality of the sound. 2. The shape of the mouth: The singer needs to pay attention to the shape of the mouth and keep it open. The proper shape of the mouth can help the sound resonate in the mouth and is conducive to the clarity and accuracy of pronunciation. Normally, the tongue should be placed at the bottom of the mouth, with the tip of the tongue lightly touching the lower gums, and the upper palate should be relaxed to ensure the maximum volume of the mouth.

[0053] Scoring criteria: 1 point: the cavity is fully opened and maintains elasticity and self-consistency; 2 points: the cavity is fully opened but not in all pitches; 3 points: the cavity is not opened enough or lacks elasticity; 4 points: the cavity is difficult to open, the muscles are stiff and not self-consistent; 5 points: the cavity is squeezed, narrow and stiff.

[0054] 5. clean

[0055] In the purity of the timbre of Bel Canto singing,

[0056] In Bel Canto, the purity of the timbre of the voice refers to the purity and clarity of the voice, as well as the degree of accurate matching with the pitch. This concept is closely related to the quality and accuracy of the voice. In Bel Canto, the pursuit of the purity of the timbre of the voice means that the singer must ensure the purity of the voice quality, without noise or broken voice through correct vocalization techniques and resonance control. This requires the singer to maintain good breathing control, ensure the stability and uniformity of the airflow, no air leakage, good closure of the vocal cords, and correctly adjust the larynx and resonance cavity so that the voice can fully resonate and resonate. In addition, the purity of the timbre of the voice also involves the accuracy and stability of the voice. The singer needs to accurately control the pitch to ensure that the voice fully matches the pitch required by the music, while maintaining the stability and coherence of the timbre and avoiding fluctuations or deviations in the pitch.

[0057] Scoring criteria: 1 point: The sound remains pure at all pitches, without air leakage, hoarseness or other rough timbre problems. 2 points: The sound can remain pure at most pitches (80%), but there are air leakage, hoarseness or other rough timbre problems at a few pitches. 3 points: The sound can remain pure at most pitches (60-70%). 4 points: The sound can remain pure at some pitches (50%). 5 points: The overall timbre of the sound is hoarse or cracked.

[0058] 6. resonate

[0059] Resonance in Bel Canto refers to the process of using the body's cavities to enhance the timbre and volume of the voice. This includes the resonance of cavities such as the mouth, throat, nasal cavity, and head and chest cavities. First, oral resonance refers to the resonance effect of the sound in the mouth. Through the correct oral shape and throat opening, the singer can make the sound resonate in the mouth, thereby enhancing the clarity and brightness of the sound. Secondly, throat resonance refers to the resonance effect of the sound in the throat cavity. The correct throat position and opening degree can help singers achieve throat resonance and make the voice fuller and more powerful. In addition, nasal resonance refers to the resonance effect of the sound in the nasopharyngeal cavity. Proper nasopharyngeal cavity opening and airflow control can help singers achieve nasal resonance and make the voice more warm and full. Finally, head cavity resonance and chest cavity resonance refer to the resonance effect of the sound in the head and chest cavities. Through correct breathing techniques and postures, singers can achieve resonance in the head and chest cavities, making the voice deeper, fuller and more powerful.

[0060] Scoring criteria: 1 point: full resonance in all ranges, 2 points: good full resonance in most ranges (80%), 3 points: good resonance in some ranges (60-70%), 4 points: good resonance only in a few ranges, 5 points: no resonance

[0061] 7. unify

[0062] Unification of voice: Bel Canto singing emphasizes the unity of voice in each voice range. The principle of unified voice range of Bel Canto singing requires that the vocalization method, voice position, timbre and volume of the three voice ranges of high, medium and low should be the same. If they are not the same, some voice ranges may sound bright while others may sound obscure. At the same time, the transition (change of voice range) between the three voice ranges of high, medium and low should be smooth without interruption, so as to achieve complete unity.

[0063] Scoring criteria: 1 point: Maintain uniform timbre in all pitches. 2 points: Uniform timbre in the three ranges of high, medium and low pitches. 3 points: The timbre is not completely uniform in the range and the transition range. 4 points: The timbre is not uniform in the transition range and the interval range. 5 points: All ranges are not uniform.

[0064] 8. falsetto

[0065] Falsetto refers to the sound effect of the edge of the vocal cords vibrating during singing. In bel canto, falsetto usually refers to those higher pitches than ordinary chest voice. The sound texture is lighter and thinner, giving people an ethereal and suspended feeling. In practical applications, the use of falsetto can achieve the expression of higher ranges without excessive effort, allowing the sound to flow freely without obvious pressure. This technique is particularly important for opera and classical music singers, who need to switch between different ranges while maintaining the beauty and expressiveness of the voice. The use of falsetto is not limited to high notes, it can also create special sound effects in the middle and bass areas.

[0066] Scoring criteria: 1 point: The sound of vocal edge vibration is sufficient, 2 points: The sound of vocal cord edge vibration is good, 3 points: The sound of vocal cord edge vibration is only reflected in certain sound ranges, 4 points: The sound of vocal cord edge vibration is less, 5 points: There is no sound of vocal cord edge vibration.

[0067] 9. chest

[0068] Chest voice refers to the sound effect of the overall vibration of the vocal cords during singing. The true voice in Bel Canto, also known as full voice or chest voice, is a singing technique based on chest resonance. When using the true voice, the singer produces sound through a more complete and powerful vocal cord vibration, relying on the resonance of the chest to enhance the power and depth of the sound. The characteristics of the true voice are that the sound is thick, full, and has a natural strength. It is the most commonly used vocalization method for singers in the middle and low range, and can produce a deep and strong timbre. In Bel Canto singing, the control of the true voice is very important because it is not only related to the expressiveness of the musical work, but also directly affects the health of the singer's voice. Correct true voice singing requires good breath support and vocal cord control to avoid oppression and damage to the vocal cords. In performance practice, singers usually use the true voice when they need to express emotional intensity or in the climax of the song to achieve a more moving and shocking effect. At the same time, the effective conversion between true voice and false voice is also an important skill in advanced singing techniques.

[0069] Scoring criteria: 1 point: The overall vibration sound of the vocal cords is sufficient, 2 points: The overall vibration sound of the vocal cords is good, 3 points: The overall vibration sound of the vocal cords is only reflected in certain pitch ranges, 4 points: The overall vibration of the vocal cords is less, 5 points: There is no overall vibration sound of the vocal cords.

[0070] 10.nasal

[0071] The nasal timbre is a wrong timbre in singing, which will cause the singing timbre to be dull, unclear, poor in resonance, poor in penetration, and other problems. We often use whether the singer's singing is accompanied by a nasal timbre to judge the singer's singing quality.

[0072] Scoring criteria: 1 point: No nasal tone at all, 2 points: Some pitches have nasal problems (less than 10%). 3 points: A large number of notes have nasal problems (less than 30%). 4 points: Most pitches have nasal problems (more than half). 5 points: All pitches have nasal problems.

[0073] In a further design, the Audio Spectrogram Transformer (AST) model is a convolution-free, purely attention-based model that can be directly applied to audio spectrograms and can capture the global content of the spectrum even at the lowest level. AST has superior performance and outperforms state-of-the-art systems on a variety of audio classification tasks and datasets. At the same time, AST also supports variable-length inputs and can be applied to different tasks without changing the architecture. Compared with convolutional neural networks, AST has a simpler structure, fewer parameters, and faster convergence in the training process. It is the first audio classification model based purely on attention.

[0074] AST works as follows: First, the input audio waveform of t seconds is converted into a 128-dimensional sequence of log-Mel filter bank (fbank) features, computed every 10ms using a 25ms Hamming window. This results in a 128×100t spectrogram as the input to AST. We then split the spectrogram into a sequence of N 16×16 patches with an overlap of 6 in both the time and frequency dimensions, where N=12d(100t-16)=10e is the number of patches and the effective input sequence length of the transformer. We use a linear projection layer to flatten each 16×16 patch into a 1D patch embedding of size 768. We call this linear projection layer the patch embedding layer.

[0075] Since the Transformer architecture cannot capture input order information and the patch sequence is not temporal, we add a trainable position embedding (also of size 768) to each patch embedding to allow the model to capture the spatial structure of the 2D audio spectrogram.

[0076] In the present invention, the audio data is loaded using the fast resampling method in the Librosa library, and then converted into MFCC (Mel-frequency cepstral coefficients), and this step is used as a unified feature extraction and connected to the machine learning algorithm for scoring prediction. Mel-frequency cepstral coefficients (MFCC) are widely used in audio signal processing and analysis, especially in tasks such as speech recognition, music genre classification, and sound event detection. Inspired by the sensitivity of the human auditory system to different frequencies, MFCC provides a compact representation of the short-term power spectrum of sound. The main motivation behind it is to capture the basic features of audio signals related to human auditory perception, while traditional methods (such as using raw audio waveforms or spectrograms) may not be able to fully represent the important features of perception.

[0077] Training and Evaluation

[0078] In the present invention, two machine learning models of supervised learning methods are trained to verify the consistency of the data set, and the accuracy of the model is verified on a data set of the singer's sound performance before and after being verified by traditional algorithms such as spectral analysis.

[0079] The data set used in the present invention contains more than 2,000 audio segments. The audio data is converted into MFCC (Mel-frequency cepstral coefficients) as a fixed feature extractor. The AST model is then used to perform a classification task of scoring and predicting the data, and supervised learning is performed using data labels scored by professionals.

[0080] Figure 2 The data set distribution of each skill item in the Sopran10T dataset used to train the AST model is shown. It can be seen that the vast majority of the materials are concentrated in the range of 2 to 3 points, which is consistent with the actual singing level of singers in the field of vocal education. This shows that the scoring labels in the Sopran10T dataset are highly accurate and the consistency of the labels is well maintained in most techniques.

[0081] Figure 3 The accuracy and loss function of the AST model training are shown. It can be observed that as the training time increases, the accuracy of the model approaches 1 (the actual accuracy is about 96%), while the loss approaches 0. This shows that our model is very accurate, the score is extremely accurate, and the skills can be better distinguished.

[0082] Figure 4The accuracy and loss function of the AST model test are shown. It can be observed that the test accuracy of the model is above 91%, which is high, while the loss is only about 0.3. This shows that when facing a large amount of data testing, our model has high accuracy and accurate scoring, and can be used in vocal teaching.

[0083] Figure 5 The confusion matrix of the AST model under various dataset allocations in the case of 1-5 points is shown. It can be observed that the accuracy of 1 point (best) and 5 points (worst) still maintains the best level. At the same time, due to the merger of 2 and 3 points with the highest recall rate, the accuracy in the two middle segments increases. This shows that the accuracy and consistency of the labels of the ten soprano bel canto singing techniques of the Sopran10T dataset in the four-category case are at a good level.

[0084] In the present invention, the hardware deployment is as follows: Raspberry Pi is an open source hardware development board. Using Raspberry Pi as a development platform has the characteristics of convenience, integration and stability. Integrating the AST model on the Raspberry Pi development board can make the entire scoring system modular and productized.

[0085] In summary, in the present invention, the frequency spectrum of the singer's singing audio is analyzed by artificial intelligence technology. Compared with the popular singing scoring software on the market, it does not need to be compared with the original song, and has a wider scope of application. Secondly, the scoring standard of this system refers to the scoring standard of the International Bel Canto Competition, and judges the singer's singing level from the dimensions of ten technical indicators. Compared with the previous scoring standards, it is more refined and objective. Furthermore, the algorithm and model used in the present invention are extremely accurate after being trained with a large number of high-quality training sets, reaching an accuracy of more than 90%. The previous model accuracy for Bel Canto singing was only about 30-40%. This has laid a good foundation for the subsequent scoring and comparison of the differences before and after warm-up. Finally, through quantitative means, the singing effects before and after warm-up can be accurately and efficiently compared from a mathematical level, which can effectively assist vocal teachers in training and adjusting teaching plans, and promote the development of vocal education to a more advanced level.

[0086] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The Bel Canto Singing Scoring System based on the AST model is characterized by: The specific steps include: Step 1: Quantification of Bel Canto singing skills: First, the singer's singing works are divided into ten fingers, and the singer's singing data set is collected; Step 2: For these ten indicators, the international bel canto competition scoring method is used, with a five-point system, where 1 is the best, 2 is better, 3 is average, 4 is poor, and 5 is the worst; Step 3: After the data set is collected, it is scored by famous experts in the bel canto industry to ensure the accuracy of the scoring; Step 4: AI scoring and analysis of singing works: Use AI algorithms to build a model, convert the sound into a spectrum, and input it into the model for training; Step 5: Using the AST model, through training of multiple data sets, the characteristics of each indicator data are extracted to achieve accurate scoring of bel canto singing works, and the scoring results are displayed using Kiviat graphs; Step six, after scoring, use neural network as an analysis tool to analyze the changes in scores of the same singer singing the same piece at different times, analyze their correlation, compare the quality of their previous and subsequent performances, and quantitatively analyze their performance changes.

2. The Bel Canto scoring system based on the AST model according to claim 1, characterized in that: In step one, the singer's performance is divided into ten indicators, and the ten indicators are: Vibrato, Throat, Position, Open, Clean, Resonate, Unify, Falsetto, Chest, and Nasal.

3. The application of the Bel Canto scoring system based on the AST model is characterized by: Used in singing scoring software.

Images