Real-time evaluation method and system for singing pitch and resonance quality in vocal music teaching

Abstract

This invention relates to the field of vocal music teaching technology, specifically to a method and system for real-time evaluation of pitch and resonance quality in vocal music teaching. The method includes: acquiring student singing audio signals and their environmental noise characteristics through an audio acquisition device; performing adaptive noise suppression, signal gain calibration, and dynamic framing processing based on note type on the singing audio signals; synchronously inputting the pre-processed audio signals into a pitch analysis channel and a resonance evaluation channel for parallel asynchronous feature extraction; and simultaneously using a vocal state modeling module to collect initial note features to establish an individual vocal baseline. This invention achieves independent analysis and correlated evaluation of pitch and resonance quality through a parallel asynchronous dual-channel feature extraction architecture. Combined with an adaptive adjustment mechanism for the individual vocal baseline, it solves the problems in traditional vocal music teaching where teachers cannot fully cover students' after-class practice and where incorrect vocal habits are difficult to correct in a timely manner.

Description

Technical Field

[0001] This invention relates to the field of vocal music teaching technology, specifically to a method and system for real-time evaluation of vocal performance pitch and resonance quality. Background Technology

[0002] In vocal performance, pitch accuracy and resonance quality are the core elements determining the sound quality, directly affecting the professionalism and artistic expression of the performance. Pitch accuracy refers to the degree to which the sung pitch matches the standard pitch, while resonance is the sound effect created by the vibration of the vocal cords through the body's cavities. Different resonance states result in different sound textures and penetrating power. In acoustic analysis, pitch can be quantified through fundamental frequency characteristics, while resonance quality can be characterized by features such as the frequency, bandwidth, and energy distribution of formants. These acoustic characteristics can directly reflect the singer's vocal state.

[0003] In current vocal music teaching, core problems such as pitch deviations and improper resonance placement that students encounter during vocal exercises and performances mainly rely on teachers to listen to and correct them one by one. However, teachers have limited energy and cannot cover all students' after-class practice. As a result, problems such as pitch deviations and improper resonance placement cannot be corrected in a timely manner during students' independent practice after class. These incorrect vocal habits will continue to solidify without guidance, not only affecting the improvement of students' singing level, but also potentially causing damage to the vocal cords and other vocal organs due to incorrect vocal techniques, thereby affecting the overall effectiveness and efficiency of vocal music teaching. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for real-time evaluation of pitch and resonance quality in vocal music teaching. This solves the problems of traditional teaching relying on teachers' on-site listening and identification, failing to cover after-class practice, making it difficult to correct pitch and resonance problems in a timely manner, resulting in low teaching efficiency and the risk of developing incorrect vocal habits that damage the vocal cords.

[0005] To achieve the above objectives, this invention provides the following technical solution: a method for real-time evaluation of vocal performance pitch and resonance quality, comprising:

[0006] The audio signal of the student's singing and its environmental noise characteristics are acquired by an audio acquisition device, and the singing audio signal is subjected to adaptive noise suppression, signal gain calibration and dynamic frame segmentation based on note type.

[0007] The preprocessed audio signal is synchronously input into the pitch analysis channel and the resonance evaluation channel for parallel asynchronous feature extraction. At the same time, the vocal state modeling module is used to collect initial note features to establish an individual vocal baseline.

[0008] Within the pitch analysis channel, a joint fundamental frequency estimation strategy is used to extract the fundamental frequency profile curve, and the fundamental frequency profile curve is dynamically time-normalized and aligned with the reference pitch sequence of the standard piece. The pitch deviation is then calculated using a note weighting mechanism.

[0009] The frequencies, bandwidths, and energy values ​​of multiple resonance peaks are extracted within the resonance evaluation channel. After normalization of the resonance peak energy, the resonance characteristics are compared with a preset sound reference model.

[0010] A multi-feature weighted fusion algorithm is used to perform correlation analysis between pitch analysis conclusions and resonance assessment conclusions, identify the causal relationship between pitch deviation and improper resonance position, and adaptively adjust the assessment threshold according to the individual vocal baseline to generate a comprehensive assessment result.

[0011] The comprehensive evaluation results are fed back through visual charts and voice prompts, and the practice data is synchronized to the teacher's terminal to form a closed-loop teaching system.

[0012] Furthermore, in the dynamic framing process, the system adjusts the frame length and frame shift parameters in real time based on the characteristics of the singing pitch, the note type, and the singing speed.

[0013] For fast passing notes, a short frame processing mode is used to capture transient pitch changes, while for long, steady-state notes, a long frame processing mode is used to extract stable resonance characteristics, thus completely preserving the subtle pitch fluctuations during the singing process in both the time and frequency domains.

[0014] Furthermore, the pitch analysis channel adopts a combined strategy of autocorrelation fundamental frequency estimation and cepstral method, and introduces a fundamental frequency anomaly correction mechanism. The fundamental frequency profile of the performance is extracted through sliding window smoothing and outlier removal.

[0015] When calculating pitch deviation, differentiated weights are assigned based on the difficulty coefficient and artistic importance of the notes in the piece, with higher weights assigned to notes in the high register and ornaments than to sustained steady-state notes.

[0016] Furthermore, the resonance evaluation channel extracts the parameters of the first four resonance peaks through linear predictive analysis combined with short-time Fourier transform;

[0017] The frequency ratio of the first and second resonant peaks is used to analyze the degree of oral cavity opening and tongue position. The energy attenuation of the third resonant peak relative to the fundamental frequency is used to assess the sufficiency of head cavity resonance. The energy concentration of the resonant peaks in the high-frequency range is used to characterize the penetrating power of the sound.

[0018] Furthermore, the resonance feature comparison step involves introducing a dynamic learning algorithm to fine-tune the ideal range of the voice reference model in real time according to the student's voice characteristics, vocal foundation, and practice progress.

[0019] The real-time extracted formant features are compared with the calibrated reference model in multiple dimensions, and the data analysis of the vocal state modeling module is combined with the correlation between resonance deviation and specific vocal actions.

[0020] Furthermore, the vocal state modeling module receives real-time feature data of pitch and resonance, and establishes a dynamically updated personal vocal state model in combination with the difficulty of the practice pieces.

[0021] The vocal cord fatigue state is determined by identifying the fundamental frequency shift trend and the stability of the resonance position, and the state correction coefficient is provided for subsequent feature fusion calibration based on the vocal state model.

[0022] Furthermore, when performing correlation analysis, the multi-feature weighted fusion algorithm determines whether the low pitch is induced by improper resonance position by judging the synchronicity between pitch abnormality and resonance deviation.

[0023] The feature fusion calibration module corrects the calculation bias of a single channel based on the results of the correlation analysis, and dynamically adjusts the confidence level of the evaluation conclusion according to the magnitude of the bias and the frequency of occurrence.

[0024] Furthermore, the step of providing feedback through visual charts and voice prompts includes:

[0025] The display terminal shows the pitch deviation by superimposing the fundamental frequency curve and the standard pitch line, and uses different colors and brightness to mark the deviation range;

[0026] The two-dimensional vowel space diagram using formant frequencies is used to show the direction of deviation of the vocal position from the ideal range, while simultaneously outputting graded voice prompts and vocal adjustment techniques.

[0027] Furthermore, the teaching closed-loop linkage steps include: automatically generating a practice evaluation report after the practice is completed, which includes a statistical histogram of pitch deviation, a trend curve of resonance quality score, and an analysis of the causes of vocal problems;

[0028] The practice evaluation report is pushed to the instructor's terminal through the data storage module, and the instructor's terminal sends back personalized practice tasks and correction comments to realize the cyclical evaluation of practice effectiveness.

[0029] This invention also provides a real-time evaluation system for vocal teaching, assessing pitch accuracy and resonance quality, including:

[0030] The audio preprocessing module connects to a professional condenser microphone and is used to perform noise suppression, gain calibration, and dynamic framing on the acquired vocal audio signal.

[0031] The pitch analysis channel, connected to the audio preprocessing module, is used to extract the fundamental frequency profile and calculate the weighted pitch deviation through a joint fundamental frequency estimation strategy.

[0032] The resonance evaluation channel, connected to the audio preprocessing module, is used to extract formant features and evaluate resonance quality in conjunction with a dynamically calibrated voice reference model.

[0033] The vocal state modeling module is connected to the pitch analysis channel and the resonance evaluation channel respectively, and is used to establish an individual vocal baseline and identify abnormal vocal states.

[0034] The feature fusion calibration module, connected to the vocal state modeling module, is used to generate a comprehensive evaluation conclusion through a multi-feature weighted fusion algorithm.

[0035] An adaptive feedback module, connected to the feature fusion calibration module, is used to output visual charts, voice prompts, and practice evaluation reports in real time.

[0036] The teacher terminal is bidirectionally connected to the adaptive feedback module and is used to receive the practice assessment report and issue personalized guidance tasks.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] This invention utilizes a parallel and asynchronous dual-channel feature extraction architecture to simultaneously achieve independent analysis and correlated evaluation of pitch and resonance quality. Combined with an adaptive adjustment mechanism for individual vocal baselines, it solves the problems in traditional vocal teaching where teachers cannot comprehensively cover students' after-class practice and where incorrect vocal habits are difficult to correct in a timely manner. The system can capture subtle pitch fluctuations and formant feature changes during singing in real time. Employing differentiated note weighting and multi-dimensional formant comparison methods, it accurately quantifies pitch deviations and multi-dimensional resonance characteristics, identifies the causal relationship between pitch shifts and improper resonance placement, and dynamically calibrates evaluation standards based on students' vocal characteristics, vocal foundation, and practice progress, avoiding errors caused by uniform evaluation standards. Simultaneously, the system provides students with immediate singing guidance through visual charts and tiered voice prompts, helping them quickly correct incorrect vocal movements and preventing vocal cord damage from improper vocal techniques. After practice, a complete evaluation report is automatically generated and synchronized to the teacher's terminal. Teachers can then provide personalized guidance tasks based on students' practice data, enabling cyclical evaluation of practice effectiveness and improving the overall efficiency and quality of vocal teaching. Attached Figure Description

[0039] Figure 1 This is a flowchart of the method of the present invention;

[0040] Figure 2 This is a flowchart illustrating the comprehensive evaluation and feature fusion calibration process of the present invention;

[0041] Figure 3 This is a flowchart illustrating the closed-loop teaching process of the present invention.

[0042] Figure 4 This is a system module diagram of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figures 1-3 This invention provides a method for real-time evaluation of vocal accuracy and resonance quality in vocal teaching, including:

[0045] The audio signal of the student's singing and its environmental noise characteristics are acquired by an audio acquisition device, and the singing audio signal is subjected to adaptive noise suppression, signal gain calibration and dynamic frame segmentation based on note type.

[0046] The preprocessed audio signal is synchronously input into the pitch analysis channel and the resonance evaluation channel for parallel asynchronous feature extraction. At the same time, the vocal state modeling module is used to collect initial note features to establish an individual vocal baseline.

[0047] Within the pitch analysis channel, a joint fundamental frequency estimation strategy is used to extract the fundamental frequency profile curve, and the fundamental frequency profile curve is dynamically time-normalized and aligned with the reference pitch sequence of the standard piece. The pitch deviation is then calculated using a note weighting mechanism.

[0048] The frequencies, bandwidths, and energy values ​​of multiple resonance peaks are extracted within the resonance evaluation channel. After normalization of the resonance peak energy, the resonance characteristics are compared with a preset sound reference model.

[0049] A multi-feature weighted fusion algorithm is used to perform correlation analysis between pitch analysis conclusions and resonance assessment conclusions, identify the causal relationship between pitch deviation and improper resonance position, and adaptively adjust the assessment threshold according to the individual vocal baseline to generate a comprehensive assessment result.

[0050] The comprehensive evaluation results are fed back through visual charts and voice prompts, and the practice data is synchronized to the teacher's terminal to form a closed-loop teaching system.

[0051] This solution employs a parallel and asynchronous dual-channel feature extraction architecture, simultaneously enabling independent analysis and correlation assessment of pitch accuracy and resonance quality. Combined with an adaptive adjustment mechanism for individual vocal baselines, it addresses the issues in traditional vocal teaching where teachers cannot fully cover students' after-class practice and where incorrect vocal habits are difficult to correct in a timely manner. It provides students with real-time and precise singing guidance and enhances overall teaching efficiency through a closed-loop teaching mechanism.

[0052] In one specific embodiment, during the dynamic framing process, the system adjusts the frame length and frame shift parameters in real time based on the characteristics of the singing pitch, the type of note, and the singing speed.

[0053] For fast passing notes, a short frame processing mode is used to capture transient pitch changes, while for long, steady-state notes, a long frame processing mode is used to extract stable resonance characteristics, thus completely preserving the subtle pitch fluctuations during the singing process in both the time and frequency domains.

[0054] Specifically, the system automatically determines the note type and singing tempo by identifying the start and end points of notes in the audio signal and combining this with note information from a standard piece of music. The frame length for fast passing notes ranges from 10ms to 20ms, and the frame shift ranges from 5ms to 10ms. This parameter setting ensures the system can capture millisecond-level transient pitch changes. The frame length for long, steady-state notes ranges from 30ms to 50ms, and the frame shift ranges from 15ms to 25ms. Longer frame lengths improve frequency domain resolution, resulting in more stable and accurate formant features. The system switches frame processing modes in real-time during performance without manual intervention, ensuring high-quality audio features are obtained in various singing scenarios.

[0055] In one specific embodiment, the pitch analysis channel adopts a joint strategy of autocorrelation fundamental frequency estimation and cepstral method, and introduces a fundamental frequency anomaly correction mechanism. The singing fundamental frequency profile is extracted through sliding window smoothing and outlier removal.

[0056] When calculating pitch deviation, differentiated weights are assigned based on the difficulty coefficient and artistic importance of the notes in the piece, with higher weights assigned to notes in the high register and ornaments than to sustained steady-state notes.

[0057] Specifically, the autocorrelation fundamental frequency estimation method is robust to periodic signals, while the cepstral method can effectively separate the fundamental frequency from the formant information. Combining these two methods improves the accuracy of fundamental frequency estimation. The fundamental frequency anomaly correction mechanism employs a sliding window smoothing process, with the window size ranging from 5 to 10 frames. Noise interference is smoothed by calculating the average fundamental frequency value within the window. Outlier removal uses the 3σ criterion, identifying and removing fundamental frequency values ​​exceeding three standard deviations from the average value as outliers.

[0058] The formula for calculating pitch accuracy score is derived from a common quantitative method for pitch accuracy assessment in vocal music teaching. This application improves upon this by introducing a note weighting mechanism. The pitch deviation of a single note is the absolute difference between the average fundamental frequency of the actual singing of that note and the standard reference fundamental frequency. The weighted pitch deviation is the sum of the products of the pitch deviations of all notes and their corresponding weighting coefficients, divided by the sum of all weighting coefficients. The formula for calculating pitch accuracy score is as follows:

[0059]

[0060] in, The score is for pitch analysis, with the dimension being points; The total number of musical notes; For the first The weighting coefficient of each note, dimensionless; For the first The average fundamental frequency of each note, measured in Hz; For the first The standard reference fundamental frequency for each musical note is measured in Hz.

[0061] The derivation of the formula is as follows: First, calculate the absolute deviation between the actual singing fundamental frequency and the standard reference fundamental frequency for each note. Then, assign different weights based on the note's difficulty and artistic importance, calculate the weighted average deviation, and finally subtract the weighted average deviation from the maximum score of 100 to obtain the pitch score. In the note weighting mechanism, the weight range for high-pitched notes is 1.5 to 2.0, for ornaments it is 1.2 to 1.8, and for sustained steady-state notes it is 0.8 to 1.0. The weight is positively correlated with the note's difficulty and artistic importance; the higher the difficulty and the stronger the artistic expression of the note, the greater its weight in the pitch assessment.

[0062] Calculation Example: A student sings a melody containing 3 notes, where the first note is a sustained steady-state segment, with weights... Actual base frequency Standard fundamental frequency The second note is a high-pitched note, with weight. Actual base frequency Standard fundamental frequency The third note is an ornament, with weight. Actual base frequency Standard fundamental frequency .

[0063] Substitute into the formula to calculate:

[0064]

[0065] Therefore, the student's pitch score for this melody is 97.52.

[0066] In one specific embodiment, the resonance evaluation channel extracts the parameters of the first four resonance peaks through linear predictive analysis combined with short-time Fourier transform;

[0067] The frequency ratio of the first and second resonant peaks is used to analyze the degree of oral cavity opening and tongue position. The energy attenuation of the third resonant peak relative to the fundamental frequency is used to assess the sufficiency of head cavity resonance. The energy concentration of the resonant peaks in the high-frequency range is used to characterize the penetrating power of the sound.

[0068] Specifically, linear predictive analysis can accurately estimate the formant frequencies of the vocal tract, while short-time Fourier transform provides detailed spectral energy distribution information. Combining the two allows for the extraction of more comprehensive and accurate formant parameters. The first formant typically ranges from 200Hz to 1000Hz, primarily reflecting the degree of oral cavity opening; higher frequencies indicate greater oral cavity opening. The second formant typically ranges from 800Hz to 2500Hz, primarily reflecting the anterior-posterior position of the tongue; higher frequencies indicate a more forward tongue position. The third formant typically ranges from 2000Hz to 3500Hz, and its energy attenuation is negatively correlated with the fullness of head cavity resonance; smaller attenuation indicates more full head cavity resonance. The energy concentration of the formants in the high-frequency range is obtained by calculating the proportion of energy in the 3000Hz to 5000Hz frequency band to the total energy; a higher proportion indicates stronger sound penetration.

[0069] The formula for calculating the resonance score originates from the quantitative evaluation method of formant characteristics in acoustic analysis. This application, based on this method, combines multi-dimensional formant characteristics for comprehensive calculation. The formula for calculating the resonance score is as follows:

[0070]

[0071] in, The resonance assessment score is based on points. The characteristic scores for the first and second resonance peaks are given, with the dimension being points. The characteristic score for the third resonance peak is expressed in units of 1. The score represents the energy concentration of the high-frequency resonance peak, with dimensions of points.

[0072] The derivation of the formula is as follows: Resonance quality is decomposed into three main dimensions: oral cavity opening and tongue position, head cavity resonance sufficiency, and sound penetration. Different weights are assigned to each dimension based on their influence on resonance quality. Then, the scores of the three dimensions are weighted and summed to obtain the total resonance score. The scores for each dimension are calculated using a deviation normalization method, with the specific formula as follows: ,in For the feature scores of the corresponding dimension, For real-time extracted feature values, This represents the ideal value for this feature in the acoustic reference model. This represents the maximum allowable deviation value for this feature. Feature scores for the first and second resonance peaks. The first formant frequency deviation score is the arithmetic mean of the second formant frequency deviation scores; the third formant characteristic score is... Calculation based on the energy attenuation of the third resonant relative to the fundamental frequency; energy concentration score of the high-frequency resonant. The energy percentage is calculated based on the frequency band from 3000Hz to 5000Hz. The scoring method for each dimension is as follows: the real-time extracted feature values ​​are compared with the ideal values ​​in the sound reference model, and the score is calculated according to the degree of deviation. The smaller the deviation, the higher the score.

[0073] Calculation Example: A male baritone student sings the vowel "a". Feature scores of the first formant and the second formant. Score, characteristic score of the third resonance peak Score for high-frequency resonance peak energy concentration point.

[0074] Substitute into the formula to calculate:

[0075]

[0076] Therefore, the student's resonance score for singing the vowel "a" was 90.1 points.

[0077] In one specific embodiment, the resonance feature comparison step involves introducing a dynamic learning algorithm to fine-tune the ideal range of the voice reference model in real time based on the student's voice characteristics, vocal foundation, and practice progress.

[0078] The real-time extracted formant features are compared with the calibrated voice reference model in multiple dimensions, and the data analysis of the vocal state modeling module is combined with the correlation between resonance deviation and specific vocal actions.

[0079] Specifically, the initial establishment of the vocal reference model is based on a large amount of acoustic data from professional singers, divided into different vocal parts such as soprano, mezzo-soprano, tenor, and baritone. Each vocal part contains ideal formant ranges corresponding to different vowels. The dynamic learning algorithm gradually adjusts the ideal ranges of the reference model by collecting students' historical practice data, making the evaluation results more closely match the students' actual vocal abilities. The system compares the real-time extracted formant frequencies, bandwidths, and energy values ​​with the calibrated reference model, calculating the deviation value for each dimension. Combined with data analysis from the vocal state modeling module, the system can map resonance deviations to specific vocal movements, such as insufficient oral cavity opening, tongue position too far back, and insufficient soft palate elevation, providing students with targeted adjustment suggestions.

[0080] In one specific embodiment, the vocal state modeling module receives real-time characteristic data of pitch and resonance, and establishes a dynamically updated personal vocal state model in combination with the difficulty of the practice pieces.

[0081] The vocal cord fatigue state is determined by identifying the fundamental frequency shift trend and the stability of the resonance position, and the state correction coefficient is provided for subsequent feature fusion calibration based on the vocal state model.

[0082] Specifically, when students use the system for the first time, the system will guide them to sing multiple audio signals with different pitches and vowels, collect initial vocal characteristics, and establish a personal vocal baseline.

[0083] The system guides students to sing five standard vowels and eight long notes of different pitches, each lasting more than three seconds, covering their vocal range. The system collects the fundamental frequency range, frequency and energy distribution characteristics of the first four formants for each note. Outlier removal and average calculation are performed on the collected data to obtain the student's basic vocal parameters in a relaxed state, which serve as their individual vocal baseline.

[0084] The individual vocal state model is a dynamic parameter model built upon an individual vocal baseline to characterize the changes in vocal characteristics of students at different practice stages and under different fatigue states. The model uses the individual vocal baseline as the initial value and continuously collects data on pitch deviation, resonance characteristic deviation, fundamental frequency stability, and formant stability during each practice session, dynamically updating the model using a sliding window averaging method. The model maintains a feature sequence containing data from the most recent 20 practice sessions. After each practice session, the average feature value of that session is added to the sequence, the oldest feature value is removed, the mean and standard deviation of the sequence are recalculated, and the model parameters are updated.

[0085] During each practice session, the vocal state modeling module receives real-time feature data from the pitch analysis and resonance assessment channels, dynamically updating the individual's vocal state model in conjunction with the difficulty level of the practice piece. The system determines vocal cord fatigue by analyzing the fundamental frequency shift trend and resonance position stability of multiple consecutive notes. When the fundamental frequency shift gradually increases and the resonance position stability decreases, vocal cord fatigue is identified. At this point, the system automatically adjusts subsequent assessment thresholds and prompts the student to take appropriate rest to avoid overuse of the voice and subsequent vocal cord damage.

[0086] The vocal state correction score can be used to quantify the impact of vocal cord fatigue on singing performance. The formula for calculating the vocal state correction score is as follows:

[0087]

[0088] in, The score is adjusted based on vocalization status, with the dimension being points; This is a state correction coefficient, dimensionless, with a value range of 0.1 to 0.5; It represents the average change in pitch deviation over 10 consecutive notes, with the dimension Hz; The average change in resonance deviation over 10 consecutive notes is expressed in minutes.

[0089] The derivation of this formula is as follows: Calculate the changes in pitch deviation and resonance deviation for multiple consecutive notes. The sum of these two reflects the degree of vocal cord fatigue. Then, multiply by a state correction coefficient to obtain a fatigue deduction. Finally, subtract the fatigue deduction from the maximum score of 100 to obtain the vocal state correction score. State correction coefficient. Adjust the settings according to the difficulty level of the practice pieces; the higher the difficulty, the better. The larger the value, the better.

[0090] Calculation Example: The change in average pitch deviation of 10 consecutive notes during a student's practice. Average resonance deviation change The state correction coefficient k = 0.2.

[0091] Substitute into the formula to calculate:

[0092]

[0093] Therefore, the student's vocalization correction score at this time is 98.9 points.

[0094] In a specific embodiment, when performing correlation analysis, the multi-feature weighted fusion algorithm determines whether the low pitch is induced by improper resonance position by judging the synchronicity between pitch abnormality and resonance deviation.

[0095] The feature fusion calibration module corrects the calculation bias of a single channel based on the results of the correlation analysis, and dynamically adjusts the confidence level of the evaluation conclusion according to the magnitude of the bias and the frequency of occurrence.

[0096] Specifically, the core of the multi-feature weighted fusion algorithm is to integrate information from three dimensions: pitch, resonance, and vocalization state, to generate a comprehensive and accurate evaluation result. The formula for calculating the comprehensive evaluation score, proposed in this application, is used to fuse multiple independent evaluation dimensions into a single comprehensive evaluation index. The formula for calculating the comprehensive evaluation score is as follows:

[0097]

[0098] in, To comprehensively evaluate the score, the unit of measurement is points; The score is for pitch analysis, with the dimension being points; The resonance assessment score is based on points. The score is adjusted for vocalization state, with the dimension being points; α, β, and γ are weighting coefficients, which are dimensionless and satisfy α+β+γ=1.

[0099] The derivation of this formula is as follows: Different weights are assigned based on the influence of pitch, resonance, and vocalization on the singing effect. Then, the scores from the three dimensions are weighted and summed to obtain a comprehensive evaluation score. The initial weights are set to α=0.5, β=0.4, and γ=0.1. The system dynamically adjusts the weight coefficients based on the student's practice. For students with significant pitch problems, the weights are appropriately increased. The value of [value]; for students with more prominent resonance problems, appropriately increase [the value]. The value of .

[0100] The logical relationship between the three formulas above is as follows: First, calculate the pitch score, resonance score, and vocal state correction score respectively. Then, substitute these three scores into the comprehensive evaluation score formula to obtain the final comprehensive evaluation result.

[0101] Calculation Example: Using the calculation results of the scores for each dimension mentioned above, the pitch score is calculated as follows. Score, resonance score Score for vocalization correction The initial weights are α=0.5, β=0.4, and γ=0.1.

[0102] Substitute into the formula to calculate:

[0103] S=0.5×97.52+0.4×90.1+0.1×98.9=48.76+36.04+9.89=94.69;

[0104] Therefore, the student's overall evaluation score for this performance was 94.69.

[0105] The system performs correlation analysis by judging the temporal synchronicity between pitch abnormalities and resonance deviations. If a low pitch is accompanied by an increased energy attenuation of the third formant, it indicates that the low pitch is caused by insufficient head cavity resonance. In this case, the feature fusion calibration module corrects the calculation deviation of the pitch analysis channel, reduces the deduction for the pitch deviation of that note, and adjusts the confidence level of the evaluation conclusion. The larger the deviation and the higher the frequency of occurrence, the higher the confidence level of the evaluation conclusion.

[0106] In one specific embodiment, the step of providing feedback through visual charts and voice prompts includes: displaying the pitch deviation on the display terminal through a superimposed comparison chart of the fundamental frequency curve and the standard pitch line, and marking the deviation magnitude using different colors and brightness.

[0107] The two-dimensional vowel space diagram using formant frequencies is used to show the direction of deviation of the vocal position from the ideal range, while simultaneously outputting graded voice prompts and vocal adjustment techniques.

[0108] Specifically, in the superimposed comparison diagram of the fundamental frequency curve and the standard pitch line, the standard pitch line is represented by a solid black line, while the fundamental frequency curve sung by the student is represented by a solid colored line. The system uses different colors to mark the pitch deviation based on its magnitude: deviations within ±5 cents are represented by green, deviations between ±5 and ±20 cents by yellow, and deviations exceeding ±20 cents by red. The brightness of the color is positively correlated with the deviation magnitude; the larger the deviation, the brighter the color. The two-dimensional vowel space diagram of formant frequencies plots the ideal ranges for different vowels with the first formant frequency as the horizontal axis and the second formant frequency as the vertical axis. The student's real-time vocal position is represented by blue dots, visually showing the direction and distance of deviation from the ideal range. Voice prompts are divided into three levels, each corresponding to a different deviation magnitude. The system synchronously outputs specific vocal adjustment techniques to help students quickly correct incorrect vocal movements.

[0109] In one specific embodiment, the teaching closed-loop linkage steps include: automatically generating a practice evaluation report after the practice is completed, which includes a statistical histogram of pitch deviation, a trend curve of resonance quality score, and an analysis of the causes of vocal problems;

[0110] The practice evaluation report is pushed to the instructor's terminal through the data storage module, and the instructor's terminal sends back personalized practice tasks and correction comments to realize the cyclical evaluation of practice effectiveness.

[0111] Specifically, practice assessment reports are automatically generated after each practice session, requiring no manual intervention. A histogram of pitch deviation statistics displays the distribution of notes with different deviation ranges, helping students understand the distribution of their pitch problems. A resonance quality score trend curve shows the changes in resonance quality during practice, helping students observe their progress. The vocal problem analysis section summarizes the main problems students encounter and provides targeted adjustment suggestions. Practice assessment reports are uploaded to a cloud server via a data storage module and automatically pushed to the corresponding instructor's terminal device. Teachers can view detailed student practice data, including the pitch deviation and resonance characteristics of each note. Based on the student's practice, teachers send back personalized practice tasks and feedback, which students receive on their terminals and practice accordingly. The system records the student's practice process and progress, forming a closed-loop teaching system of cyclical assessment, continuously improving teaching effectiveness.

[0112] Please see Figure 4 The present invention also provides a real-time evaluation system for vocal teaching, including:

[0113] The audio preprocessing module connects to a professional condenser microphone and is used to perform noise suppression, gain calibration, and dynamic framing on the acquired vocal audio signal.

[0114] The pitch analysis channel, connected to the audio preprocessing module, is used to extract the fundamental frequency profile and calculate the weighted pitch deviation through a joint fundamental frequency estimation strategy.

[0115] The resonance evaluation channel, connected to the audio preprocessing module, is used to extract formant features and evaluate resonance quality in conjunction with a dynamically calibrated voice reference model.

[0116] The vocal state modeling module is connected to the pitch analysis channel and the resonance evaluation channel respectively, and is used to establish an individual vocal baseline and identify abnormal vocal states.

[0117] The feature fusion calibration module, connected to the vocal state modeling module, is used to generate a comprehensive evaluation conclusion through a multi-feature weighted fusion algorithm.

[0118] An adaptive feedback module, connected to the feature fusion calibration module, is used to output visual charts, voice prompts, and practice evaluation reports in real time.

[0119] The teacher terminal is bidirectionally connected to the adaptive feedback module and is used to receive the practice assessment report and issue personalized guidance tasks.

[0120] Specifically, the audio preprocessing module connects to a professional condenser microphone with a sampling rate of 44.1kHz and a sampling precision of 16-bit, enabling the acquisition of high-quality audio signals. The pitch analysis channel and resonance evaluation channel are connected in parallel to the audio preprocessing module for synchronous feature extraction, ensuring real-time evaluation. The vocal state modeling module connects to both the pitch analysis and resonance evaluation channels, receiving feature data in real time and updating individual vocal state models. The feature fusion calibration module receives the output from the vocal state modeling module and the analysis results from both channels, generating a comprehensive evaluation conclusion through a multi-feature weighted fusion algorithm. The adaptive feedback module converts the comprehensive evaluation conclusion into visual charts and voice prompts, providing real-time feedback to students and generating a practice evaluation report after practice. The teacher terminal connects bidirectionally to the adaptive feedback module via a wireless network, enabling data transmission and task distribution, supporting simultaneous online practice for multiple students and unified teacher management.

[0121] In summary, this invention achieves independent analysis and correlated evaluation of pitch and resonance quality through a parallel and asynchronous dual-channel feature extraction architecture. Combined with an adaptive adjustment mechanism for individual vocal baselines, it solves the problems in traditional vocal teaching where teachers cannot fully cover students' after-class practice and where incorrect vocal habits are difficult to correct in a timely manner. The system can capture subtle pitch fluctuations and formant feature changes during singing in real time. Employing differentiated note weighting and multi-dimensional formant comparison methods, it accurately quantifies pitch deviations and multi-dimensional resonance characteristics, identifies the causal relationship between pitch shifts and improper resonance placement, and dynamically calibrates evaluation standards based on students' vocal characteristics, vocal foundation, and practice progress, avoiding errors caused by uniform evaluation standards. Simultaneously, the system provides students with immediate singing guidance through visual charts and tiered voice prompts, helping them quickly correct incorrect vocal movements and preventing damage to the vocal cords from improper vocalization. After practice, a complete evaluation report is automatically generated and synchronized to the teacher's terminal. Teachers can then provide personalized guidance tasks based on students' practice data, enabling cyclical evaluation of practice effectiveness and improving the overall efficiency and quality of vocal teaching.

[0122] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time evaluation of pitch accuracy and resonance quality in vocal music teaching, characterized in that, include: The audio signal of the student's singing and its environmental noise characteristics are acquired by an audio acquisition device, and the singing audio signal is subjected to adaptive noise suppression, signal gain calibration and dynamic frame segmentation based on note type. The pre-processed singing audio signal is synchronously input into the pitch analysis channel and resonance evaluation channel for parallel asynchronous feature extraction. At the same time, the vocal state modeling module is used to collect initial note features to establish an individual vocal baseline. Within the pitch analysis channel, a joint fundamental frequency estimation strategy is used to extract the fundamental frequency profile curve, and the fundamental frequency profile curve is dynamically time-normalized and aligned with the reference pitch sequence of the standard piece. The pitch deviation is then calculated using a note weighting mechanism. The frequencies, bandwidths, and energy values ​​of multiple resonance peaks are extracted within the resonance evaluation channel. After normalization of the resonance peak energy, the resonance characteristics are compared with a preset sound reference model. A multi-feature weighted fusion algorithm is used to perform correlation analysis between pitch analysis conclusions and resonance assessment conclusions, identify the causal relationship between pitch deviation and improper resonance position, and adaptively adjust the assessment threshold according to the individual vocal baseline to generate a comprehensive assessment result. The comprehensive evaluation results are fed back through visual charts and voice prompts, and the practice data is synchronized to the teacher's terminal to form a closed-loop teaching system.

2. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, In the dynamic framing process, the frame length and frame shift parameters are adjusted in real time according to the characteristics of the singing pitch, the type of note, and the singing speed. For fast passing tones, a short frame processing mode is used to capture transient pitch changes, while for long steady-state segments, a long frame processing mode is used to extract stable resonance features.

3. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, The pitch analysis channel employs a combined strategy of autocorrelation fundamental frequency estimation and cepstral method, and introduces a fundamental frequency anomaly correction mechanism. It extracts the singing fundamental frequency profile through sliding window smoothing and outlier removal. When calculating pitch deviation, differentiated weights are assigned based on the difficulty coefficient and artistic importance of the notes in the piece, with higher weights assigned to notes in the high register and ornaments than to sustained steady-state notes.

4. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, The resonance evaluation channel extracts the parameters of the first four resonance peaks through linear predictive analysis combined with short-time Fourier transform; The frequency ratio of the first and second resonant peaks is used to analyze the degree of oral cavity opening and tongue position. The energy attenuation of the third resonant peak relative to the fundamental frequency is used to assess the sufficiency of head cavity resonance. The energy concentration of the resonant peaks in the high-frequency range is used to characterize the penetrating power of the sound.

5. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, The resonance feature comparison step uses a dynamic learning algorithm to fine-tune the ideal range of the voice reference model in real time according to the student's voice characteristics, vocal foundation and practice progress. The real-time extracted formant features are compared with the calibrated voice reference model in multiple dimensions, and the data analysis of the vocal state modeling module is combined with the correlation between resonance deviation and specific vocal actions.

6. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, The vocal state modeling module receives real-time characteristic data of pitch and resonance, and establishes a dynamically updated personal vocal state model based on the difficulty of the practice pieces. The vocal cord fatigue state is determined by identifying the fundamental frequency shift trend and the stability of the resonance position, and the state correction coefficient is provided for subsequent feature fusion calibration based on the vocal state model.

7. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, When performing correlation analysis, the multi-feature weighted fusion algorithm determines whether the low pitch is induced by improper resonance position by judging the synchronicity between pitch abnormality and resonance deviation. The feature fusion calibration module corrects the calculation bias of a single channel based on the results of the correlation analysis, and dynamically adjusts the confidence level of the evaluation conclusion according to the magnitude of the bias and the frequency of occurrence.

8. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, The steps for providing feedback through visual charts and voice prompts include: The display terminal shows the pitch deviation by superimposing the fundamental frequency curve and the standard pitch line, and uses different colors and brightness to mark the deviation range; The two-dimensional vowel space diagram using formant frequencies is used to show the direction of deviation of the vocal position from the ideal range, while simultaneously outputting graded voice prompts and vocal adjustment techniques.

9. The method for real-time evaluation of vocal performance pitch and resonance quality according to claim 1, characterized in that, The teaching closed-loop linkage steps include: automatically generating a practice evaluation report after the practice is completed, which includes a histogram of pitch deviation statistics, a trend curve of resonance quality score, and an analysis of the causes of vocal problems; The practice assessment report is pushed to the instructor's terminal through the data storage module, and the instructor's terminal then sends back the personalized practice tasks and feedback.

10. A real-time evaluation system for vocal teaching performance pitch and resonance quality, used to execute the real-time evaluation method for vocal teaching performance pitch and resonance quality as described in any one of claims 1-9, characterized in that, include: The audio preprocessing module connects to a professional condenser microphone and is used to perform noise suppression, gain calibration, and dynamic framing on the acquired vocal audio signal. The pitch analysis channel, connected to the audio preprocessing module, is used to extract the fundamental frequency profile and calculate the weighted pitch deviation through a joint fundamental frequency estimation strategy. The resonance evaluation channel, connected to the audio preprocessing module, is used to extract formant features and evaluate resonance quality in conjunction with a dynamically calibrated voice reference model. The vocal state modeling module is connected to the pitch analysis channel and the resonance evaluation channel respectively, and is used to establish an individual vocal baseline and identify abnormal vocal states. The feature fusion calibration module, connected to the vocal state modeling module, is used to generate a comprehensive evaluation conclusion through a multi-feature weighted fusion algorithm. An adaptive feedback module, connected to the feature fusion calibration module, is used to output visual charts, voice prompts, and practice evaluation reports in real time. The teacher terminal is bidirectionally connected to the adaptive feedback module and is used to receive the practice assessment report and issue personalized guidance tasks.

Images