Comparison and teaching demonstration method and system for vocal music performance style characteristics

Abstract

The present application relates to the technical field of vocal music teaching, and discloses a vocal music work singing style feature comparison and teaching demonstration method and system, the method comprising: firstly, performing heterogeneous granularity segmentation on the demonstration singing audio and the student singing audio of a target vocal music work to obtain a demonstration sentence segmentation sequence and a student audio frame sequence; secondly, performing multi-dimensional feature decomposition on the demonstration audio sentence to obtain a demonstration style feature parameter set; then, analyzing the habit stability by analyzing the feature dimension fluctuation to obtain a style demonstration benchmark by weighted fusion; matching the student audio frame sequence by using the benchmark to obtain a comparison sentence pair; measuring the multi-dimensional deviation between the real-time style feature parameters and the demonstration style feature parameters to generate a style difference vector; finally, projecting the difference vector into a visual double-track comparison map, and obtaining a student priority practice sentence set through saliency discrimination; the present application can improve the accuracy of vocal music singing style feature comparison and the pertinence of vocal music teaching demonstration.

Description

Technical Field

[0001] This invention relates to the field of vocal music teaching technology, and in particular to a method and system for comparing and demonstrating the characteristics of vocal performance styles. Background Technology

[0002] Current methods for comparing vocal performance style characteristics rely on subjective human judgment, making it impossible to accurately differentiate between demonstration and student performances at a granular level. This hinders the complete extraction of multi-dimensional style characteristics such as breath control, articulation, dynamics, vibrato, and timbre. Furthermore, the comparison results lack unified standards and objective evidence, compromising accuracy. Existing vocal teaching demonstrations do not establish standardized style benchmarks based on the stable characteristics of demonstration performances. The content of teaching demonstrations is disconnected from students' actual performance, failing to provide students with a standardized reference for appropriate practice.

[0003] Current technologies cannot achieve precise, phrase-level constraint matching between student singing audio and demonstration singing audio, making it difficult to quantify deviations in multi-dimensional style characteristics, generate intuitive visual comparison charts, or clearly present the trajectory of differences in singing styles. Existing teaching methods do not select key practice content based on the degree of style deviation, teaching guidance lacks clear priorities, students cannot specifically correct singing deviations, and the efficiency of vocal style optimization and the effectiveness of teaching demonstrations fail to meet practical needs. Therefore, improving the accuracy of vocal style characteristic comparison and the targeted nature of vocal teaching demonstrations has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method and system for comparing and demonstrating the characteristics of vocal performance styles, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for comparing and demonstrating the stylistic features of vocal works, comprising:

[0006] C1. Perform heterogeneous granular segmentation on the demonstration singing audio and student singing audio of the target vocal work to obtain the demonstration phrase segmentation sequence and student audio frame sequence of the target vocal work.

[0007] C2. Perform multi-dimensional feature deconstruction on the demonstration audio phrases in the segmented sequence of the demonstration musical phrases to obtain the demonstration style feature parameter set of the demonstration audio phrases;

[0008] C3. Perform volatility quantification analysis on the feature dimensions of the demonstration style feature parameter set to obtain the habitual stability of the feature dimensions, and based on the habitual stability, perform weighted fusion on the demonstration style feature parameter set to obtain the style demonstration benchmark of the target vocal work.

[0009] C4. Based on the style demonstration benchmark, perform constraint matching on the student audio frame sequence to obtain the matching musical phrase pairs of the student audio frame sequence;

[0010] C5. Perform a multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the demonstration style feature parameter set to obtain the style difference vector of the compared musical phrase pair.

[0011] C6. Project the deviation trajectory of the style difference vector to obtain a dual-track comparison graph of the style difference vector on the visualization interface, and perform a saliency judgment on the style difference vector to obtain the set of student priority practice phrases for the target vocal work.

[0012] In a preferred embodiment, the heterogeneous granularity segmentation of the demonstration singing audio and the student singing audio of the target vocal work to obtain the demonstration phrase segmentation sequence and the student audio frame sequence of the target vocal work includes:

[0013] Obtain demonstration and student recordings of the target vocal works;

[0014] Amplitude envelope extraction is performed on the demonstration singing audio to obtain the local energy valley points of the demonstration singing audio;

[0015] Based on the local energy valley points, boundary determination is performed on the demonstration singing audio to obtain the demonstration phrase segmentation sequence of the target vocal work;

[0016] A sliding window is used to capture the student's singing audio to obtain the student audio frame sequence of the target vocal work.

[0017] In a preferred embodiment, the step of performing multi-dimensional feature deconstruction on the demonstration audio phrases in the segmented sequence of the demonstration musical phrases to obtain the demonstration style feature parameter set of the demonstration audio phrases includes:

[0018] A time-frequency joint analysis is performed on the example audio phrases in the segmented sequence of the example musical phrases to obtain the energy envelope, fundamental frequency trajectory, and spectral distribution sequence of the example audio phrases;

[0019] Steady-state evaluation of the energy envelope is performed to obtain the breath support stability of the energy envelope. Multimodal feature fusion is then performed on the breath support stability, the onset mode of the initial segment and the finish mode of the tail segment in the energy envelope to obtain the breath application feature parameters of the demonstration audio phrase.

[0020] The syllables in the lyrics of the example audio phrase are subjected to auxiliary transition detection to obtain the articulation feature parameters of the example audio phrase;

[0021] The energy envelope is converted to decibel scale to obtain the dynamic dynamic change curve of the demonstration audio phrase, and the trend feature is extracted from the dynamic dynamic change curve to obtain the dynamic performance feature parameters of the demonstration audio phrase.

[0022] The frequency periodic fluctuations of the long note segment in the fundamental frequency trajectory are demodulated using time-varying frequency to obtain the fluctuation amplitude and fluctuation rate of the fundamental frequency trajectory, and the fluctuation amplitude and fluctuation rate are used as the vibrato characteristic parameters of the example audio phrase.

[0023] The spectral distribution sequence is subjected to power-weighted statistics to obtain the spectral centroid and spectral tilt of the spectral distribution sequence, and the spectral centroid and spectral tilt are used as the timbre change feature parameters of the demonstration audio phrase;

[0024] By integrating the breath control feature parameters, the articulation feature parameters, the dynamic performance feature parameters, the vibrato feature parameters, and the timbre variation feature parameters, a set of demonstration style feature parameters for the demonstration audio phrase is obtained.

[0025] In a preferred embodiment, the step of performing volatility quantification analysis on the feature dimensions of the demonstration style feature parameter set to obtain the habitual stability of the feature dimensions, and then performing weighted fusion on the demonstration style feature parameter set based on the habitual stability to obtain the style demonstration benchmark of the target vocal work, includes:

[0026] The feature dimensions of the demonstration style feature parameter set are extracted across musical phrases to obtain the cross-musical phrase value sequence of the feature dimensions;

[0027] The difference between two adjacent musical phrases in the cross-phrase value sequence is statistically analyzed to obtain the average fluctuation amplitude of the feature dimension, and the probability quality assessment of the peak frequency range in the cross-phrase value sequence is performed to obtain the concentration ratio of the feature dimension.

[0028] By performing a joint parameter mapping between the average fluctuation amplitude and the concentration ratio, the habitual stability of the feature dimension is obtained.

[0029] Based on the habit stability, habit-oriented weighting is applied to the demonstration audio phrases in the demonstration style feature parameter set to obtain the weighted feature vector of the demonstration style feature parameter set;

[0030] Based on the temporal order of the sample audio phrases, the weighted feature vectors are aggregated into temporal vectors to obtain the stylistic demonstration benchmark of the target vocal work.

[0031] In a preferred embodiment, the formula for calculating the habit stability is as follows:

[0032] ;

[0033] In the formula, For the aforementioned habit stability, The concentration ratio is mentioned. The average fluctuation amplitude, The range of the value sequence across musical phrases. The interquartile range of the value sequence across musical phrases. It is a natural constant. The preset concentration power adjustment parameter, This is the preset fluctuation amplitude scaling parameter.

[0034] In a preferred embodiment, the step of performing constraint matching on the student audio frame sequence based on the style demonstration benchmark to obtain the matched phrase pairs of the student audio frame sequence includes:

[0035] Based on the exemplary audio phrases in the style demonstration benchmark, the temporal position index is performed on the segmented sequence of the exemplary audio phrases to obtain the temporal anchor point position of the exemplary audio phrases in the exemplary singing audio.

[0036] The weighted feature vector in the style demonstration benchmark is used as the matching benchmark for the student audio frame sequence;

[0037] Based on the time anchor position, the real-time feature vector of the student audio frame sequence is correlated with the matching benchmark to obtain the student audio frame with the minimum distance to the demonstration audio phrase, and the position of the student audio frame with the minimum distance is used as the candidate matching point of the demonstration audio phrase.

[0038] Based on the order of the sample audio phrases, the candidate matching points are arranged sequentially to obtain the candidate matching point sequence.

[0039] Based on the candidate matching point sequence, the student audio frame sequence is truncated by anchor point intervals to obtain the matching phrase pairs of the student audio frame sequence.

[0040] In a preferred embodiment, the step of performing a multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the exemplary style feature parameter set to obtain the style difference vector of the compared musical phrase pair includes:

[0041] The difference between the onset of the real-time breath application feature parameters in the compared musical phrase pair and the onset of the demonstrated breath application feature parameters in the demonstration style feature parameter set is evaluated to obtain the breath deviation of the compared musical phrase pair.

[0042] The articulation deviation of the compared musical phrases is obtained by measuring the temporal distance between the auxiliary component ratio sequence of the real-time articulation feature parameters in the compared musical phrases and the auxiliary component ratio sequence of the articulation feature parameters in the demonstration style feature parameter set.

[0043] The dynamic dynamic change curve of the real-time dynamic performance characteristic parameter in the compared musical phrase pair and the demonstration dynamic change curve of the demonstration dynamic performance characteristic parameter set of the demonstration style characteristic parameter are accumulated point by point to obtain the dynamic deviation of the compared musical phrase pair.

[0044] The absolute difference between the real-time fluctuation amplitude of the dynamic dynamic change curve and the exemplary fluctuation amplitude of the exemplary dynamic change curve is used as the amplitude deviation of the compared musical phrase pair, and the absolute difference between the real-time fluctuation rate of the dynamic dynamic change curve and the exemplary fluctuation rate of the exemplary dynamic change curve is used as the rate deviation of the compared musical phrase pair.

[0045] By concatenating the breath deviation, articulation deviation, dynamic deviation, amplitude deviation, and speed deviation in a multidimensional series, the style difference vector of the compared musical phrase pair is obtained.

[0046] In a preferred embodiment, the deviation trajectory projection of the style difference vector is performed to obtain a dual-track comparison graph of the style difference vector on the visualization interface, and the saliency of the style difference vector is determined to obtain the student priority practice phrase set of the target vocal work, including:

[0047] The style difference vector is subjected to inverse bias mapping to obtain the demonstration feature parameters of the demonstration audio phrase in the feature dimension. The phrase number of the demonstration audio phrase and the demonstration feature parameters are then subjected to scatter linear interpolation to obtain the demonstration trajectory curve of the style difference vector.

[0048] The style difference vector is reconstructed by residual additive reconstruction to obtain the real-time feature parameters of the student audio frame sequence in the feature dimension, and the musical phrase number is concatenated with the real-time feature parameters by coordinate sequence to obtain the student trajectory curve of the style difference vector.

[0049] The demonstration trajectory curve and the student trajectory curve are simultaneously superimposed and plotted to obtain a dual-track comparison map of the style difference vector in the visualization interface;

[0050] Multidimensional deviation reduction is performed on the style difference vector to obtain the comprehensive deviation value of the compared musical phrase pair;

[0051] Based on the comprehensive deviation value, the compared musical phrase pairs are filtered to obtain the set of student-priority practice phrases for the target vocal work.

[0052] In a preferred embodiment, the formula for calculating the comprehensive deviation value is as follows:

[0053] ;

[0054] In the formula, The comprehensive deviation value is... The onset deviation of the compared musical phrase pairs is marked. The breath deviation of the compared musical phrase pair is the amount of breath control. The pronunciation deviation of the compared musical phrase pair is the amount of the pronunciation deviation. The dynamic deviation of the compared musical phrase pair is given. The deviation in vibrato amplitude of the compared musical phrase pair. The deviation in vibrato rate of the compared musical phrase pair. It is a natural constant.

[0055] To address the aforementioned problems, this invention also provides a vocal performance style feature comparison and teaching demonstration system, the system comprising:

[0056] The heterogeneous granularity segmentation module is used to perform heterogeneous granularity segmentation on the demonstration singing audio and student singing audio of the target vocal work to obtain the demonstration musical phrase segmentation sequence and student audio frame sequence of the target vocal work.

[0057] The multi-dimensional feature deconstruction module is used to perform multi-dimensional feature deconstruction on the demonstration audio phrases in the segmented sequence of the demonstration musical phrases to obtain the demonstration style feature parameter set of the demonstration audio phrases;

[0058] The fluctuation-weighted fusion module is used to perform fluctuation quantification analysis on the feature dimensions of the demonstration style feature parameter set, obtain the habitual stability of the feature dimensions, and perform weighted fusion on the demonstration style feature parameter set based on the habitual stability to obtain the style demonstration benchmark of the target vocal work.

[0059] The constraint matching alignment module is used to perform constraint matching on the student audio frame sequence based on the style demonstration benchmark to obtain the matching phrase pairs of the student audio frame sequence.

[0060] The multidimensional deviation measurement module is used to perform multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the demonstration style feature parameter set to obtain the style difference vector of the compared musical phrase pair.

[0061] The deviation projection discrimination module is used to project the deviation trajectory of the style difference vector to obtain a dual-track comparison map of the style difference vector in the visualization interface, and to perform saliency discrimination on the style difference vector to obtain the set of student priority practice phrases for the target vocal work.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. This invention can completely deconstruct the multi-dimensional style features of demonstration singing audio, accurately quantify the habitual stability of feature dimensions and complete weighted fusion to form a standardized style demonstration benchmark, realize precise constraint matching between student singing audio and demonstration singing audio, fully measure style differences in dimensions such as breath, articulation, dynamics, vibrato, and timbre, and ensure the objectivity and accuracy of singing style feature comparison.

[0064] 2. This invention can transform stylistic differences into intuitive dual-track comparison charts, and complete the significance judgment based on the deviation value to lock in the priority practice phrases, making teaching demonstrations more targeted, directly guiding students to correct singing deviations in a targeted manner, improving the efficiency of singing style optimization, giving vocal teaching guidance a clear execution priority, and continuously strengthening the implementation effect of teaching demonstrations and the efficiency of improving singing ability. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating a method for comparing and demonstrating the characteristics of vocal performance styles in an embodiment of the present invention.

[0066] Figure 2 This is a functional module diagram of a vocal performance style feature comparison and teaching demonstration system provided in an embodiment of the present invention;

[0067] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0068] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0069] This application provides a method for comparing and demonstrating the performance style characteristics of vocal works. The execution subject of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0070] Reference Figure 1 The diagram shown is a flowchart illustrating a method for comparing and demonstrating the singing style characteristics of vocal works according to an embodiment of the present invention. In this embodiment, the method includes:

[0071] C1. Perform heterogeneous granular segmentation on the demonstration singing audio and student singing audio of the target vocal work to obtain the demonstration phrase segmentation sequence and student audio frame sequence of the target vocal work.

[0072] In this embodiment of the invention, the heterogeneous granularity segmentation of the demonstration singing audio and the student singing audio of the target vocal work to obtain the demonstration phrase segmentation sequence and the student audio frame sequence of the target vocal work includes:

[0073] Obtain demonstration and student recordings of the target vocal works;

[0074] Amplitude envelope extraction is performed on the demonstration singing audio to obtain the local energy valley points of the demonstration singing audio;

[0075] Based on the local energy valley points, boundary determination is performed on the demonstration singing audio to obtain the demonstration phrase segmentation sequence of the target vocal work;

[0076] A sliding window is used to capture the student's singing audio to obtain the student audio frame sequence of the target vocal work.

[0077] According to the preset audio acquisition standards, the demonstration performance audio and student performance audio of the target vocal works are collected and organized to form complete audio materials that can be used for subsequent processing.

[0078] The audio acquisition standards are as follows: both the demonstration singing audio and the student singing audio should be recorded in mono, with a sampling precision of no less than 16 bits, and the sampling frequency should be set to a value that can cover the fundamental frequency and main overtone range of the human voice. The audio file format should be an uncompressed waveform file format.

[0079] The signal amplitude of the demonstration singing audio is collected point by point in the time domain. The signal amplitude at consecutive time points is connected to form a complete amplitude change profile, and the amplitude envelope is extracted. The position where the signal amplitude drops to the preset minimum energy threshold is located on the amplitude envelope, and the local energy valley point of the demonstration singing audio is obtained.

[0080] Using local energy valleys as the boundaries for segmenting musical phrases in the demonstration singing audio, the demonstration singing audio is segmented according to the chronological order of the boundary points. The segmented audio segments are then arranged in chronological order to obtain the segmented sequence of the demonstration musical phrases of the target vocal work.

[0081] A segmented window of fixed length moves segment by segment across the entire time domain of the student's singing audio. Each fixed time interval between movements completes one segmentation of the audio content within the window. All the segmented audio segments are arranged sequentially according to the movement order to obtain the student audio frame sequence of the target vocal work.

[0082] By using differentiated segmentation methods, regular demonstration phrase segmentation sequences and student audio frame sequences are obtained respectively, which can adapt to the audio feature extraction needs of demonstration singing and student singing, providing stable and accurate audio basic units for subsequent style feature deconstruction and matching alignment, and improving the coherence and reliability of the overall processing flow.

[0083] C2. Perform multi-dimensional feature deconstruction on the demonstration audio phrases in the segmented sequence of the demonstration musical phrases to obtain the demonstration style feature parameter set of the demonstration audio phrases;

[0084] In this embodiment of the invention, the step of performing multi-dimensional feature deconstruction on the demonstration audio phrases in the segmented sequence of the demonstration musical phrases to obtain the demonstration style feature parameter set of the demonstration audio phrases includes:

[0085] A time-frequency joint analysis is performed on the example audio phrases in the segmented sequence of the example musical phrases to obtain the energy envelope, fundamental frequency trajectory, and spectral distribution sequence of the example audio phrases;

[0086] Steady-state evaluation of the energy envelope is performed to obtain the breath support stability of the energy envelope. Multimodal feature fusion is then performed on the breath support stability, the onset mode of the initial segment and the finish mode of the tail segment in the energy envelope to obtain the breath application feature parameters of the demonstration audio phrase.

[0087] The syllables in the lyrics of the example audio phrase are subjected to auxiliary transition detection to obtain the articulation feature parameters of the example audio phrase;

[0088] The energy envelope is converted to decibel scale to obtain the dynamic dynamic change curve of the demonstration audio phrase, and the trend feature is extracted from the dynamic dynamic change curve to obtain the dynamic performance feature parameters of the demonstration audio phrase.

[0089] The frequency periodic fluctuations of the long note segment in the fundamental frequency trajectory are demodulated using time-varying frequency to obtain the fluctuation amplitude and fluctuation rate of the fundamental frequency trajectory, and the fluctuation amplitude and fluctuation rate are used as the vibrato characteristic parameters of the example audio phrase.

[0090] The spectral distribution sequence is subjected to power-weighted statistics to obtain the spectral centroid and spectral tilt of the spectral distribution sequence, and the spectral centroid and spectral tilt are used as the timbre change feature parameters of the demonstration audio phrase;

[0091] By integrating the breath control feature parameters, the articulation feature parameters, the dynamic performance feature parameters, the vibrato feature parameters, and the timbre variation feature parameters, a set of demonstration style feature parameters for the demonstration audio phrase is obtained.

[0092] For each segment of the demonstration audio phrase in the segmented sequence, the amplitude data of the signal is continuously collected in the time domain at fixed time intervals to form a continuously changing amplitude curve. In the frequency domain, the frequency components of the signal are decomposed at fixed frequency intervals and the energy distribution of each component is recorded. The time domain amplitude change and the frequency domain component distribution are combined and analyzed to obtain the energy envelope, fundamental frequency trajectory and spectral distribution sequence of the demonstration audio phrase.

[0093] Along the entire trajectory of the energy envelope, the duration for which the amplitude fluctuation range remains within the preset stable amplitude range of vocal breath is statistically analyzed. This stable duration is divided by the total duration of the demonstration audio phrase to obtain the breath support stability of the energy envelope. The breath support stability, the onset mode of rapid rise or gradual entry of amplitude in the initial segment of the energy envelope, and the finish mode of smooth drop or truncated amplitude in the final segment of the energy envelope are combined and correlated according to characteristic dimensions to obtain the breath application characteristic parameters of the demonstration audio phrase.

[0094] Each syllable of the corresponding lyrics within the sample audio phrase is located frame by frame. The time intervals for consonant and vowel pronunciation within the syllable are divided. The duration, smoothness, and breakpoints of the transition from the consonant to the vowel pronunciation interval are detected. Based on the detection results, feature data in a fixed format is generated to obtain the articulation feature parameters of the sample audio phrase.

[0095] According to the standard decibel conversion rules for vocal audio, the original linear amplitude values ​​of the energy envelope are converted one by one into decibel values ​​on a logarithmic scale. The converted continuous decibel values ​​are arranged in chronological order to form a curve, thus obtaining the dynamic dynamic change curve of the demonstration audio phrase. The rising rate, duration of stable maintenance, and falling rate morphological characteristics of the curve are extracted according to the beginning, middle, and end stages of the phrase, thus obtaining the dynamic performance characteristic parameters of the demonstration audio phrase.

[0096] In the fundamental frequency trajectory, continuous segments whose duration reaches the preset minimum duration threshold of a long note are selected. The frequency changes in these segments are analyzed point by point. The maximum and minimum differences of the periodic frequency changes are extracted to obtain the fluctuation amplitude. The number of times the frequency completes a periodic change per unit time is extracted to obtain the fluctuation rate. The fluctuation amplitude and fluctuation rate are combined into fixed-format feature data, which are used as the vibrato feature parameters of the demonstration audio phrase.

[0097] The energy values ​​corresponding to each frequency point in the spectral distribution sequence are weighted and accumulated. The center frequency position of the spectral energy concentration is determined by the energy value as the weight, and the spectral centroid is obtained. The difference in the proportion of energy between the high frequency band and the low frequency band is statistically analyzed to determine the overall tilt direction and tilt degree of the spectrum, and the spectral tilt degree is obtained. The spectral centroid and the spectral tilt degree are combined into fixed format feature data, which are used as the timbre change feature parameters of the demonstration audio phrase.

[0098] The parameters of breath control, articulation, dynamics, vibrato, and timbre variation are collected and organized in a fixed dimensional order to form a complete and dimensionally unified set of feature data, thus obtaining the set of feature parameters of the demonstration audio phrase.

[0099] By conducting a comprehensive and detailed feature deconstruction of the sample audio phrases, five core singing style features—breath control, articulation, dynamics, vibrato, and timbre—are fully extracted, forming a standardized and reusable set of sample style feature parameters. This provides a comprehensive, accurate, and unified feature basis for subsequent style benchmark construction, student singing feature comparison, and teaching guidance, thereby improving the completeness and accuracy of singing style analysis.

[0100] C3. Perform volatility quantification analysis on the feature dimensions of the demonstration style feature parameter set to obtain the habitual stability of the feature dimensions, and based on the habitual stability, perform weighted fusion on the demonstration style feature parameter set to obtain the style demonstration benchmark of the target vocal work.

[0101] In this embodiment of the invention, the step of performing volatility quantification analysis on the feature dimensions of the demonstration style feature parameter set to obtain the habitual stability of the feature dimensions, and then performing weighted fusion on the demonstration style feature parameter set based on the habitual stability to obtain the style demonstration benchmark of the target vocal work, includes:

[0102] The feature dimensions of the demonstration style feature parameter set are extracted across musical phrases to obtain the cross-musical phrase value sequence of the feature dimensions;

[0103] The difference between two adjacent musical phrases in the cross-phrase value sequence is statistically analyzed to obtain the average fluctuation amplitude of the feature dimension, and the probability quality assessment of the peak frequency range in the cross-phrase value sequence is performed to obtain the concentration ratio of the feature dimension.

[0104] By performing a joint parameter mapping between the average fluctuation amplitude and the concentration ratio, the habitual stability of the feature dimension is obtained.

[0105] Based on the habit stability, habit-oriented weighting is applied to the demonstration audio phrases in the demonstration style feature parameter set to obtain the weighted feature vector of the demonstration style feature parameter set;

[0106] Based on the temporal order of the sample audio phrases, the weighted feature vectors are aggregated into temporal vectors to obtain the stylistic demonstration benchmark of the target vocal work.

[0107] The formula for calculating the habit stability is as follows:

[0108] ;

[0109] In the formula, For the aforementioned habit stability, The concentration ratio is mentioned. The average fluctuation amplitude, The range of the value sequence across musical phrases. The interquartile range of the value sequence across musical phrases. It is a natural constant. The preset concentration power adjustment parameter, This is the preset fluctuation amplitude scaling parameter.

[0110] From the characteristic parameters of the demonstration style, including breath control, articulation, dynamics, vibrato, and timbre, feature values ​​of each dimension corresponding to the musical phrases in the demonstration audio are extracted. The extracted values ​​are then arranged in the order of the performance of the musical phrases in the demonstration audio to obtain the cross-phrase value sequence of the feature dimensions. This sequence provides complete data support for the calculation of all subsequent feature parameters.

[0111] Iterate through all adjacent values ​​in the cross-phrase value sequence, calculate the difference between each pair of values, sum all the calculated differences, and divide by the total number of adjacent value combinations to obtain the average fluctuation amplitude of the feature dimension, corresponding to the formula in... , is used to characterize the magnitude of the fluctuation of the feature dimension across musical phrases.

[0112] The number of values ​​in the cross-phrase value sequence that fall within a preset high-frequency feature value range is counted. This number is then divided by the total number of values ​​in the cross-phrase value sequence to obtain the concentration ratio of the feature dimension, corresponding to the formula in... It is used to characterize the degree of clustering of feature dimension values.

[0113] Extract the maximum and minimum values ​​from the cross-phrase value sequence, and subtract the minimum value from the maximum value to obtain the range of the cross-phrase value sequence, corresponding to the formula in... It is used to characterize the overall fluctuation range of the feature dimension.

[0114] Sort the cross-phrase value sequence in ascending order of numerical value, and extract the values ​​at the 25th and 75th percentiles respectively. Subtract the 25th percentile value from the 75th percentile value to obtain the interquartile range of the cross-phrase value sequence, corresponding to the formula... It is used to characterize the degree of dispersion of the middle segment of the feature dimension.

[0115] The four characteristics—average volatility, concentration ratio, range, and interquartile range—are combined with fixed natural constants in their respective formulas. And a concentration power-law adjustment parameter set based on historical statistical data of vocal performance style characteristics and vocal teaching practice experience. Fluctuation amplitude scaling parameter The habitual stability of the feature dimension is obtained by fusion calculation through preset feature stability correspondence rules, corresponding to the formula in... It is used to quantify the stability of each feature dimension of the demonstration style feature parameter set during cross-phrase singing, providing a precise basis for weight allocation in subsequent weighted fusion.

[0116] The characteristic stability rule is to multiply the average fluctuation amplitude by a fluctuation amplitude scaling parameter to obtain the scaled average fluctuation amplitude. The natural constant is then calculated. The negatively scaled average fluctuation amplitude raised to the power of the natural constant. Using the negative scaled average volatility as the base, an exponential operation is performed to obtain the exponential decay term. Subtracting the exponential decay term from the value 1 yields the numerator. Using the scaled average volatility as the denominator, the numerator is divided by the denominator to obtain the decay correction value for the average volatility. Multiplying the power-wise magnification of the concentration ratio by the decay correction value for the average volatility yields the first product. Calculating the concentration ratio divided by the sum of the value 1, the average volatility, and the concentration ratio yields the joint regulation ratio. Taking the square root of the joint regulation ratio yields the joint square root regulation value. Multiplying the first product by the joint square root regulation value yields the second product. The difference between the maximum and minimum values ​​extracted from the cross-phrase value sequence yields the range. Dividing the range by the sum of the range and the interquartile range yields the range percentage. Subtracting the range percentage from the value 1 yields the distribution dispersion correction value. Multiplying the second product by the distribution dispersion correction value yields the habitual stability of the feature dimension.

[0117] Using habit stability as the weighting standard, each feature dimension of each demonstration audio phrase in the demonstration style feature parameter set is assigned a corresponding weight according to the numerical level of habit stability. The weights are then combined with the corresponding feature values ​​to obtain the weighted feature vector of the demonstration style feature parameter set.

[0118] Based on the actual performance time sequence of the demonstration audio phrases in the target vocal work, the weighted feature vectors corresponding to all demonstration audio phrases are integrated and connected in sequence, maintaining the temporal correspondence of the weighted feature vectors, to obtain the stylistic demonstration benchmark of the target vocal work.

[0119] By analyzing the volatility across musical phrases, the stable state of each feature dimension is clarified. Combined with the habitual stability quantification calculation method that integrates multiple features, weighted fusion and temporal aggregation are completed to form a style demonstration benchmark that fits the demonstration singing habits and is standardized. This provides an accurate and reliable reference for matching and comparing student singing audio in the future.

[0120] C4. Based on the style demonstration benchmark, perform constraint matching on the student audio frame sequence to obtain the matching musical phrase pairs of the student audio frame sequence;

[0121] In this embodiment of the invention, the step of performing constraint matching on the student audio frame sequence based on the style demonstration benchmark to obtain the matching phrase pairs of the student audio frame sequence includes:

[0122] Based on the exemplary audio phrases in the style demonstration benchmark, the temporal position index is performed on the segmented sequence of the exemplary audio phrases to obtain the temporal anchor point position of the exemplary audio phrases in the exemplary singing audio.

[0123] The weighted feature vector in the style demonstration benchmark is used as the matching benchmark for the student audio frame sequence;

[0124] Based on the time anchor position, the real-time feature vector of the student audio frame sequence is correlated with the matching benchmark to obtain the student audio frame with the minimum distance to the demonstration audio phrase, and the position of the student audio frame with the minimum distance is used as the candidate matching point of the demonstration audio phrase.

[0125] Based on the order of the sample audio phrases, the candidate matching points are arranged sequentially to obtain the candidate matching point sequence.

[0126] Based on the candidate matching point sequence, the student audio frame sequence is truncated by anchor point intervals to obtain the matching phrase pairs of the student audio frame sequence.

[0127] Based on the sample audio phrases in the style demonstration benchmark, the temporal position index is carried out on the segmented sequence of sample audio phrases. The start and end times of the sample audio phrases in the sample performance audio are read segment by segment. The start and end times are combined into fixed temporal positioning identifiers to accurately mark the time interval of each sample audio phrase in the complete sample performance audio, thus obtaining the time anchor position of the sample audio phrase in the sample performance audio.

[0128] The weighted feature vector is extracted from the style demonstration benchmark after weighting based on habit stability. This weighted feature vector is set as the fixed reference standard for feature matching of student audio frame sequences. The feature comparison of all student audio frames is based solely on this standard, forming the matching benchmark for student audio frame sequences.

[0129] Using the time domain range corresponding to the time anchor point as the matching limit interval, the real-time feature vector corresponding to each frame of audio within this time domain range is extracted from the student audio frame sequence. The feature dimension distribution of each set of real-time feature vectors is compared with that of the matching benchmark. The audio frame with the highest degree of feature distribution matching is selected as the minimum distance student audio frame, where the minimum distance is measured using Euclidean distance. Specifically, for each student audio frame, the differences between the real-time feature vector and the matching benchmark vector are calculated in five dimensions: breath control, articulation, dynamics, vibrato amplitude, and vibrato rate. Each difference is squared, summed, and the square root of the sum is taken as the Euclidean distance value for that frame. Among all compared frames, the frame with the smallest Euclidean distance value is the minimum distance student audio frame.

[0130] The time position corresponding to the student audio frame with the minimum distance is determined as the candidate matching point for the current demonstration audio phrase.

[0131] According to the actual singing order of the demonstration audio phrases in the target vocal work, all candidate matching points corresponding to the demonstration audio phrases are arranged in the same singing order to maintain a one-to-one correspondence and temporal synchronization between the candidate matching points and the demonstration audio phrases, thus obtaining the candidate matching point sequence.

[0132] Based on the time position of each candidate matching point in the candidate matching point sequence, the audio extraction interval is defined with reference to the fixed duration of the corresponding demonstration audio phrase. Continuous audio content is extracted from the student audio frame sequence according to the defined interval. All extracted student audio segments are combined with the corresponding demonstration audio phrases to form a complete matching unit, resulting in the matching phrase pair of the student audio frame sequence.

[0133] By precisely locating time-domain anchor points and matching feature benchmark constraints, the temporal alignment and feature correspondence between the demonstration singing phrases and the student singing audio frames are achieved. The temporal arrangement and precise extraction of anchor point intervals form a regular and unified pair of comparison phrases, providing a standard and aligned comparison unit for subsequent multi-dimensional measurement of singing style feature deviations, effectively improving the accuracy and process stability of style difference analysis.

[0134] C5. Perform a multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the demonstration style feature parameter set to obtain the style difference vector of the compared musical phrase pair.

[0135] In this embodiment of the invention, the step of performing a multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the exemplary style feature parameter set to obtain the style difference vector of the compared musical phrase pair includes:

[0136] The difference between the onset of the real-time breath application feature parameters in the compared musical phrase pair and the onset of the demonstrated breath application feature parameters in the demonstration style feature parameter set is evaluated to obtain the breath deviation of the compared musical phrase pair.

[0137] The articulation deviation of the compared musical phrases is obtained by measuring the temporal distance between the auxiliary component ratio sequence of the real-time articulation feature parameters in the compared musical phrases and the auxiliary component ratio sequence of the articulation feature parameters in the demonstration style feature parameter set.

[0138] The dynamic dynamic change curve of the real-time dynamic performance characteristic parameter in the compared musical phrase pair and the demonstration dynamic change curve of the demonstration dynamic performance characteristic parameter set of the demonstration style characteristic parameter are accumulated point by point to obtain the dynamic deviation of the compared musical phrase pair.

[0139] The absolute difference between the real-time fluctuation amplitude of the dynamic dynamic change curve and the exemplary fluctuation amplitude of the exemplary dynamic change curve is used as the amplitude deviation of the compared musical phrase pair, and the absolute difference between the real-time fluctuation rate of the dynamic dynamic change curve and the exemplary fluctuation rate of the exemplary dynamic change curve is used as the rate deviation of the compared musical phrase pair.

[0140] By concatenating the breath deviation, articulation deviation, dynamic deviation, amplitude deviation, and speed deviation in a multidimensional series, the style difference vector of the compared musical phrase pair is obtained.

[0141] By comparing the onset of the real-time breath control parameters in the musical phrases with the onset of the demonstrated breath control parameters, the differences are judged item by item from three preset judgment dimensions: the rate of increase of the amplitude at the start of the vocalization, the duration of the start, and the smooth transition state. The degree of difference obtained from each dimension is converted into a unified quantitative value according to the preset vocal breath difference quantification standard to obtain the breath deviation of the compared musical phrases.

[0142] By comparing the auxiliary ratio sequence of the real-time articulation feature parameters in the musical phrases with the auxiliary ratio sequence of the articulation feature parameters in the demonstration style feature parameters, the numerical intervals of the corresponding positions are measured one by one according to the time scale points that are completely consistent in the two sets of sequences. The numerical intervals of all time scale points are accumulated and divided by the total number of points involved in the measurement to obtain the articulation deviation of the compared musical phrases.

[0143] By comparing the dynamic dynamic change curves of the real-time dynamic performance characteristic parameters in the musical phrases with the demonstration dynamic change curves of the demonstration style characteristic parameters, the corresponding dynamic values ​​of the two sets of curves are extracted point by point according to the same time scale, and the point position difference is calculated. The point position differences of all time scales throughout the process are continuously accumulated to obtain the dynamic deviation of the compared musical phrases.

[0144] Extract the real-time fluctuation amplitude of the dynamic dynamic change curve and the model fluctuation amplitude of the model dynamic change curve, calculate the absolute difference between the two sets of values ​​and use this difference as the amplitude deviation of the compared musical phrase pair. Extract the real-time fluctuation rate of the dynamic dynamic change curve and the model fluctuation rate of the model dynamic change curve, calculate the absolute difference between the two sets of values ​​and use this difference as the rate deviation of the compared musical phrase pair.

[0145] The deviations in breath control, articulation, dynamics, amplitude, and speed are sequentially combined and connected according to the fixed dimensions of breath control, articulation, dynamics, amplitude, and speed to form a continuous combination of deviation data with complete dimensions and consistent characteristics, thus obtaining the style difference vector of the compared musical phrases.

[0146] By conducting detailed deviation measurements on multiple dimensions of singing style characteristics, including breath control, articulation, dynamics, amplitude of fluctuation, and rate of fluctuation, a comprehensive and quantitative style difference vector is formed. This fully presents the details of the style differences between the demonstration and the student's singing, providing accurate and comprehensive quantitative data support for subsequent deviation visualization and selection of musical phrases for teaching practice.

[0147] C6. Project the deviation trajectory of the style difference vector to obtain a dual-track comparison graph of the style difference vector on the visualization interface, and perform a saliency judgment on the style difference vector to obtain the set of student priority practice phrases for the target vocal work.

[0148] In this embodiment of the invention, the deviation trajectory projection of the style difference vector is performed to obtain a dual-track comparison graph of the style difference vector on the visualization interface, and the saliency of the style difference vector is determined to obtain the student priority practice phrase set of the target vocal work, including:

[0149] The style difference vector is subjected to inverse bias mapping to obtain the demonstration feature parameters of the demonstration audio phrase in the feature dimension. The phrase number of the demonstration audio phrase and the demonstration feature parameters are then subjected to scatter linear interpolation to obtain the demonstration trajectory curve of the style difference vector.

[0150] The style difference vector is reconstructed by residual additive reconstruction to obtain the real-time feature parameters of the student audio frame sequence in the feature dimension, and the musical phrase number is concatenated with the real-time feature parameters by coordinate sequence to obtain the student trajectory curve of the style difference vector.

[0151] The demonstration trajectory curve and the student trajectory curve are simultaneously superimposed and plotted to obtain a dual-track comparison map of the style difference vector in the visualization interface;

[0152] Multidimensional deviation reduction is performed on the style difference vector to obtain the comprehensive deviation value of the compared musical phrase pair;

[0153] Based on the comprehensive deviation value, the compared musical phrase pairs are filtered to obtain the set of student-priority practice phrases for the target vocal work.

[0154] The formula for calculating the comprehensive deviation value is as follows:

[0155] ;

[0156] In the formula, The comprehensive deviation value is... The onset deviation of the compared musical phrase pairs is marked. The breath deviation of the compared musical phrase pair is the amount of breath control. The pronunciation deviation of the compared musical phrase pair is the amount of the pronunciation deviation. The dynamic deviation of the compared musical phrase pair is given. The deviation in vibrato amplitude of the compared musical phrase pair. The deviation in vibrato rate of the compared musical phrase pair. It is a natural constant.

[0157] The style difference vector is inversely mapped by deviation. According to the fixed correspondence and association rules between the style difference vector and the demonstration feature parameters, the multidimensional deviation data is reversed to restore the standard feature values ​​of the demonstration audio phrase in five feature dimensions: breath control, articulation, dynamic expression, vibrato amplitude, and vibrato rate. The original feature state of the demonstration performance is fully restored, and the demonstration feature parameters of the demonstration audio phrase in the feature dimensions are obtained. The unique sequence number of the demonstration audio phrase is used as the horizontal axis coordinate and the corresponding demonstration feature parameter value is used as the vertical axis coordinate to generate discrete scatter points. Adjacent scatter points are connected by straight lines according to the order of the phrases, and the numerical gaps between the scatter points are filled to form a continuous and smooth curve, thus obtaining the demonstration trajectory curve of the style difference vector.

[0158] The style difference vector is reconstructed by residual additive reconstruction. The deviation values ​​of each dimension within the style difference vector are superimposed point-to-point with the corresponding values ​​of the demonstration feature parameters to restore the actual singing feature values ​​of the student's audio frame sequence in the dimensions of breath control, articulation, dynamic expression, vibrato amplitude, and vibrato rate. This truly reflects the student's real-time singing feature state and obtains the real-time feature parameters of the student's audio frame sequence in the feature dimensions. Continuous coordinate points are generated with the musical phrase number, which is the same as the demonstration trajectory curve, as the horizontal axis coordinate and the real-time feature parameter values ​​as the vertical axis coordinate. All coordinate points are smoothly connected according to the singing sequence of the musical phrases to form a continuous curve without breaks, thus obtaining the student trajectory curve of the style difference vector.

[0159] In the visualization interface, a unified scale for the horizontal musical phrase number and the vertical characteristic parameter value are set to ensure that the coordinate system and measurement standard of the two curves are completely consistent. The demonstration trajectory curve and the student trajectory curve are imported into the same coordinate system and drawn synchronously. The two curves correspond to the same characteristic dimension at the same musical phrase number position, realizing the dual synchronous superposition of temporal sequence and characteristic dimension, and obtaining a dual-track comparison map of style difference vector in the visualization interface.

[0160] The multi-dimensional bias data contained in the style difference vector are integrated and calculated, whereby... The comprehensive deviation value is a quantitative indicator obtained by integrating multi-dimensional singing style deviations. It is used to characterize the degree of overall stylistic difference between the demonstration singing and the student singing in a musical phrase comparison. The onset deviation marker for the comparison phrases is derived from the evaluation results of the difference between the onset method of the real-time breath use characteristic parameters in the comparison phrases and the onset method of the demonstration breath use characteristic parameters. When the student's onset method does not match the demonstration onset method, it is marked with a preset fixed value; when there is no deviation, it is marked as 0. This is used to strengthen the weight of the impact of onset deviation on the overall deviation. The breath deviation for comparing musical phrases is derived from the evaluation results of the difference between the onset of the real-time breath application characteristic parameters and the onset of the demonstrated breath application characteristic parameters in the compared musical phrases. It is a quantitative representation of the deviation in the dimension of breath application, obtained by converting the multi-dimensional differences in the onset of the breath into a unified quantitative value. The measure of articulation deviation in the comparison of musical phrases is derived from the temporal distance measurement results of the consonant ratio sequence of real-time articulation feature parameters in the compared musical phrases and the consonant ratio sequence of the demonstration articulation feature parameters. It is a quantitative representation of the deviation in the articulation dimension, obtained by measuring the temporal interval of the two sets of consonant ratio sequences point by point and averaging them. The dynamic deviation of the compared musical phrases is derived from the cumulative residual of the dynamic dynamic change curves of the real-time dynamic performance characteristic parameters in the compared musical phrases and the demonstration dynamic change curves. It is a quantitative representation of the deviation in the dynamic performance dimension, obtained by calculating the corresponding differences between the two sets of curves point by point and summing them over the entire process. To compare the vibrato amplitude deviation of musical phrases, the absolute difference between the real-time fluctuation amplitude of the dynamic dynamic change curve and the demonstrated fluctuation amplitude of the model dynamic change curve is used. This is a quantitative representation of the deviation in the vibrato amplitude dimension, and is obtained by directly extracting the absolute difference between the two sets of fluctuation amplitudes. To compare the vibrato rate deviation of musical phrases, the absolute difference between the real-time fluctuation rate of the dynamic dynamic change curve and the exemplary fluctuation rate of the exemplary dynamic change curve is used. This is a quantitative representation of the deviation in the vibrato rate dimension, and is obtained by directly extracting the absolute difference between the two sets of fluctuation rates. This is a natural constant, derived from fixed irrational constants in mathematics. It is a pre-defined, fixed value that does not require calculation. It is used to construct an exponential deviation fusion term, achieving a non-linear quantitative representation of the deviation. This calculation method marks the initial deviation. Breath deviation Articulation deviation Force deviation vibrato amplitude deviation vibrato rate deviation This method integrates and calculates multi-dimensional singing style deviations. It enhances the impact of key singing characteristic deviations by marking onset deviations, highlights the contribution of core singing skills through the product of breath control and articulation deviations, and integrates the combined effects of dynamics and vibrato deviations through an exponential term. This achieves precise quantification of the overall singing style differences between the compared musical phrases, transforming multi-dimensional, dispersed deviation data into a single, quantifiable value that reflects the overall degree of singing style differences, thus obtaining the comprehensive deviation value for the compared musical phrase pairs. .

[0161] The overall deviation value corresponding to each pair of compared musical phrases The phrases are compared with the preset student singing priority practice judgment threshold. Phrase pairs with a comprehensive deviation value greater than or equal to the judgment threshold are selected, while phrase pairs with a comprehensive deviation value less than the threshold are removed. The selected phrase pairs are then organized and collected according to the original performance order of the work to obtain the student priority practice phrase set of the target vocal work.

[0162] By accurately reconstructing the singing characteristic trajectories of the demonstration and students through inverse deviation mapping and additive residual reconstruction, a dual-track comparison graph with unified coordinates and synchronized time sequence is generated, intuitively presenting the details of differences in singing styles, and combining a comprehensive deviation value that integrates multi-dimensional deviations. Calculation and fixed threshold filtering accurately select priority practice phrases, providing visual references and targeted practice guidance for vocal teaching, effectively improving the efficiency of singing deviation correction and the pertinence of teaching demonstrations.

[0163] like Figure 2 The diagram shown is a functional module diagram of a vocal performance style feature comparison and teaching demonstration system provided in an embodiment of the present invention.

[0164] The vocal performance style feature comparison and teaching demonstration system 100 of this invention can be installed in an electronic device. Depending on the functions implemented, the vocal performance style feature comparison and teaching demonstration system 100 may include a heterogeneous granularity segmentation module 101, a multidimensional feature deconstruction module 102, a fluctuation weighted fusion module 103, a constraint matching alignment module 104, a multidimensional deviation measurement module 105, and a deviation projection discrimination module 106. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0165] In this embodiment, the functions of each module / unit are as follows:

[0166] The heterogeneous granularity segmentation module 101 is used to perform heterogeneous granularity segmentation on the demonstration singing audio and the student singing audio of the target vocal work to obtain the demonstration musical phrase segmentation sequence and the student audio frame sequence of the target vocal work.

[0167] The multi-dimensional feature deconstruction module 102 is used to perform multi-dimensional feature deconstruction on the demonstration audio phrases in the demonstration phrase segmentation sequence to obtain the demonstration style feature parameter set of the demonstration audio phrases;

[0168] The fluctuation-weighted fusion module 103 is used to perform fluctuation quantification analysis on the feature dimensions of the demonstration style feature parameter set, obtain the habitual stability of the feature dimensions, and perform weighted fusion on the demonstration style feature parameter set based on the habitual stability to obtain the style demonstration benchmark of the target vocal work.

[0169] The constraint matching alignment module 104 is used to perform constraint matching on the student audio frame sequence based on the style demonstration benchmark to obtain the matching phrase pairs of the student audio frame sequence.

[0170] The multidimensional deviation measurement module 105 is used to perform multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the demonstration style feature parameter set to obtain the style difference vector of the compared musical phrase pair.

[0171] The deviation projection discrimination module 106 is used to project the deviation trajectory of the style difference vector to obtain a dual-track comparison map of the style difference vector on the visualization interface, and to make a saliency discrimination of the style difference vector to obtain the set of student priority practice phrases for the target vocal work.

[0172] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0173] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0174] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0175] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0176] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for comparing and demonstrating the stylistic characteristics of vocal works, characterized in that... The method includes: C1. Perform heterogeneous granular segmentation on the demonstration singing audio and student singing audio of the target vocal work to obtain the demonstration phrase segmentation sequence and student audio frame sequence of the target vocal work. C2. Perform multi-dimensional feature deconstruction on the demonstration audio phrases in the segmented sequence of the demonstration musical phrases to obtain the demonstration style feature parameter set of the demonstration audio phrases; C3. Perform volatility quantification analysis on the feature dimensions of the demonstration style feature parameter set to obtain the habitual stability of the feature dimensions, and based on the habitual stability, perform weighted fusion on the demonstration style feature parameter set to obtain the style demonstration benchmark of the target vocal work. C4. Based on the style demonstration benchmark, perform constraint matching on the student audio frame sequence to obtain the matching musical phrase pairs of the student audio frame sequence; C5. Perform a multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the demonstration style feature parameter set to obtain the style difference vector of the compared musical phrase pair. C6. Project the deviation trajectory of the style difference vector to obtain a dual-track comparison graph of the style difference vector on the visualization interface, and perform a saliency judgment on the style difference vector to obtain the set of student priority practice phrases for the target vocal work.

2. The method for comparing and demonstrating the characteristics of vocal performance styles as described in claim 1, characterized in that, The heterogeneous granularity segmentation of the demonstration singing audio and the student singing audio of the target vocal work yields the segmented sequence of the demonstration musical phrases and the sequence of student audio frames, including: Obtain demonstration and student recordings of the target vocal works; Amplitude envelope extraction is performed on the demonstration singing audio to obtain the local energy valley points of the demonstration singing audio; Based on the local energy valley points, boundary determination is performed on the demonstration singing audio to obtain the demonstration phrase segmentation sequence of the target vocal work; A sliding window is used to capture the student's singing audio to obtain the student audio frame sequence of the target vocal work.

3. The method for comparing and demonstrating the characteristics of vocal performance styles as described in claim 1, characterized in that, The multi-dimensional feature deconstruction of the demonstration audio phrases in the segmented sequence of the demonstration musical phrases yields a set of demonstration style feature parameters for the demonstration audio phrases, including: A time-frequency joint analysis is performed on the example audio phrases in the segmented sequence of the example musical phrases to obtain the energy envelope, fundamental frequency trajectory, and spectral distribution sequence of the example audio phrases; Steady-state evaluation of the energy envelope is performed to obtain the breath support stability of the energy envelope. Multimodal feature fusion is then performed on the breath support stability, the onset mode of the initial segment and the finish mode of the tail segment in the energy envelope to obtain the breath application feature parameters of the demonstration audio phrase. The syllables in the lyrics of the example audio phrase are subjected to auxiliary transition detection to obtain the articulation feature parameters of the example audio phrase; The energy envelope is converted to decibel scale to obtain the dynamic dynamic change curve of the demonstration audio phrase, and the trend feature is extracted from the dynamic dynamic change curve to obtain the dynamic performance feature parameters of the demonstration audio phrase. The frequency periodic fluctuations of the long note segment in the fundamental frequency trajectory are demodulated using time-varying frequency to obtain the fluctuation amplitude and fluctuation rate of the fundamental frequency trajectory, and the fluctuation amplitude and fluctuation rate are used as the vibrato characteristic parameters of the example audio phrase. The spectral distribution sequence is subjected to power-weighted statistics to obtain the spectral centroid and spectral tilt of the spectral distribution sequence, and the spectral centroid and spectral tilt are used as the timbre change feature parameters of the demonstration audio phrase; By integrating the breath control feature parameters, the articulation feature parameters, the dynamic performance feature parameters, the vibrato feature parameters, and the timbre variation feature parameters, a set of demonstration style feature parameters for the demonstration audio phrase is obtained.

4. The method for comparing and demonstrating the characteristics of vocal performance styles as described in claim 1, characterized in that, The process involves performing volatility quantification analysis on the feature dimensions of the demonstration style feature parameter set to obtain the habitual stability of the feature dimensions, and then, based on the habitual stability, performing weighted fusion on the demonstration style feature parameter set to obtain the style demonstration benchmark of the target vocal work, including: The feature dimensions of the demonstration style feature parameter set are extracted across musical phrases to obtain the cross-musical phrase value sequence of the feature dimensions; The difference between two adjacent musical phrases in the cross-phrase value sequence is statistically analyzed to obtain the average fluctuation amplitude of the feature dimension, and the probability quality assessment of the peak frequency range in the cross-phrase value sequence is performed to obtain the concentration ratio of the feature dimension. By performing a joint parameter mapping between the average fluctuation amplitude and the concentration ratio, the habitual stability of the feature dimension is obtained. Based on the habit stability, habit-oriented weighting is applied to the demonstration audio phrases in the demonstration style feature parameter set to obtain the weighted feature vector of the demonstration style feature parameter set; Based on the temporal order of the sample audio phrases, the weighted feature vectors are aggregated into temporal vectors to obtain the stylistic demonstration benchmark of the target vocal work.

5. The method for comparing and demonstrating the characteristics of vocal performance styles as described in claim 4, characterized in that, The formula for calculating the habit stability is as follows: ； In the formula, For the aforementioned habit stability, The concentration ratio is mentioned. The average fluctuation amplitude, The range of the value sequence across musical phrases. The interquartile range of the value sequence across musical phrases. It is a natural constant. The preset concentration power adjustment parameter, This is the preset fluctuation amplitude scaling parameter.

6. The method for comparing and demonstrating the characteristics of vocal performance styles as described in claim 2, characterized in that, The process of constraining and matching the student audio frame sequence based on the style demonstration benchmark to obtain the matched phrase pairs of the student audio frame sequence includes: Based on the exemplary audio phrases in the style demonstration benchmark, the temporal position index is performed on the segmented sequence of the exemplary audio phrases to obtain the temporal anchor point position of the exemplary audio phrases in the exemplary singing audio. The weighted feature vector in the style demonstration benchmark is used as the matching benchmark for the student audio frame sequence; Based on the time anchor position, the real-time feature vector of the student audio frame sequence is correlated with the matching benchmark to obtain the student audio frame with the minimum distance to the demonstration audio phrase, and the position of the student audio frame with the minimum distance is used as the candidate matching point of the demonstration audio phrase. Based on the order of the sample audio phrases, the candidate matching points are arranged sequentially to obtain the candidate matching point sequence. Based on the candidate matching point sequence, the student audio frame sequence is truncated by anchor point intervals to obtain the matching phrase pairs of the student audio frame sequence.

7. The method for comparing and demonstrating the characteristics of vocal performance styles as described in claim 1, characterized in that, The process of performing a multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the demonstration style feature parameter set to obtain the style difference vector of the compared musical phrase pair includes: The difference between the onset of the real-time breath application feature parameters in the compared musical phrase pair and the onset of the demonstrated breath application feature parameters in the demonstration style feature parameter set is evaluated to obtain the breath deviation of the compared musical phrase pair. The articulation deviation of the compared musical phrases is obtained by measuring the temporal distance between the auxiliary component ratio sequence of the real-time articulation feature parameters in the compared musical phrases and the auxiliary component ratio sequence of the articulation feature parameters in the demonstration style feature parameter set. The dynamic dynamic change curve of the real-time dynamic performance characteristic parameter in the compared musical phrase pair and the demonstration dynamic change curve of the demonstration dynamic performance characteristic parameter set of the demonstration style characteristic parameter are accumulated point by point to obtain the dynamic deviation of the compared musical phrase pair. The absolute difference between the real-time fluctuation amplitude of the dynamic dynamic change curve and the exemplary fluctuation amplitude of the exemplary dynamic change curve is used as the amplitude deviation of the compared musical phrase pair, and the absolute difference between the real-time fluctuation rate of the dynamic dynamic change curve and the exemplary fluctuation rate of the exemplary dynamic change curve is used as the rate deviation of the compared musical phrase pair. By concatenating the breath deviation, articulation deviation, dynamic deviation, amplitude deviation, and speed deviation in a multidimensional series, the style difference vector of the compared musical phrase pair is obtained.

8. The method for comparing and demonstrating the characteristics of vocal performance styles as described in claim 6, characterized in that, The process involves projecting the deviation trajectory of the style difference vector to obtain a dual-track comparison graph of the style difference vector on the visualization interface, and then performing a saliency judgment on the style difference vector to obtain a set of student-priority practice phrases for the target vocal work, including: The style difference vector is subjected to inverse bias mapping to obtain the demonstration feature parameters of the demonstration audio phrase in the feature dimension. The phrase number of the demonstration audio phrase and the demonstration feature parameters are then subjected to scatter linear interpolation to obtain the demonstration trajectory curve of the style difference vector. The style difference vector is reconstructed by residual additive reconstruction to obtain the real-time feature parameters of the student audio frame sequence in the feature dimension, and the musical phrase number is concatenated with the real-time feature parameters by coordinate sequence to obtain the student trajectory curve of the style difference vector. The demonstration trajectory curve and the student trajectory curve are simultaneously superimposed and plotted to obtain a dual-track comparison map of the style difference vector in the visualization interface; Multidimensional deviation reduction is performed on the style difference vector to obtain the comprehensive deviation value of the compared musical phrase pair; Based on the comprehensive deviation value, the compared musical phrase pairs are filtered to obtain the set of student-priority practice phrases for the target vocal work.

9. The method for comparing and demonstrating the stylistic features of vocal works as described in claim 8, characterized in that, The formula for calculating the comprehensive deviation value is as follows: ； In the formula, The comprehensive deviation value is... The onset deviation of the compared musical phrase pairs is marked. The breath deviation of the compared musical phrase pair is the amount of breath control. The pronunciation deviation of the compared musical phrase pair is the amount of the pronunciation deviation. The dynamic deviation of the compared musical phrase pair is given. The deviation in vibrato amplitude of the compared musical phrase pair. The deviation in vibrato rate of the compared musical phrase pair. It is a natural constant.

10. A vocal performance style comparison and teaching demonstration system, characterized in that, The system for implementing the vocal performance style feature comparison and teaching demonstration method of claim 1, the system comprising: The heterogeneous granularity segmentation module is used to perform heterogeneous granularity segmentation on the demonstration singing audio and student singing audio of the target vocal work to obtain the demonstration musical phrase segmentation sequence and student audio frame sequence of the target vocal work. The multi-dimensional feature deconstruction module is used to perform multi-dimensional feature deconstruction on the demonstration audio phrases in the segmented sequence of the demonstration musical phrases to obtain the demonstration style feature parameter set of the demonstration audio phrases; The fluctuation-weighted fusion module is used to perform fluctuation quantification analysis on the feature dimensions of the demonstration style feature parameter set, obtain the habitual stability of the feature dimensions, and perform weighted fusion on the demonstration style feature parameter set based on the habitual stability to obtain the style demonstration benchmark of the target vocal work. The constraint matching alignment module is used to perform constraint matching on the student audio frame sequence based on the style demonstration benchmark to obtain the matching phrase pairs of the student audio frame sequence. The multidimensional deviation measurement module is used to perform multidimensional deviation measurement on the real-time style feature parameter set of the compared musical phrase pair and the demonstration style feature parameter set to obtain the style difference vector of the compared musical phrase pair. The deviation projection discrimination module is used to project the deviation trajectory of the style difference vector to obtain a dual-track comparison map of the style difference vector in the visualization interface, and to perform saliency discrimination on the style difference vector to obtain the set of student priority practice phrases for the target vocal work.

Images