A musical instrument playing skill intelligent analysis method, system, device and medium

Abstract

This invention relates to the field of machine learning technology, specifically to an intelligent analysis method, system, device, and medium for musical instrument playing techniques. The method includes the following steps: collecting standard performance audio of the target instrument and typical environmental noise; fusing them to construct a priori spectral feature benchmark library; collecting live audio and filtering it to obtain a clean performance spectrum matrix; extracting multi-dimensional features and weighting them to generate a parameter set; and matching the benchmark library data to calculate and generate performance technique discrimination coefficients. In this invention, a priori spectral feature benchmark library is constructed by fusing standard performance audio and environmental noise. Constrained whitening filtering and hierarchical feature screening remove stable and sudden noises, fully preserving effective performance spectrum features directly related to playing techniques. Fine-grained characterization of playing techniques is achieved through joint weighted quantization of time-domain and frequency-domain harmonic multi-dimensional features. Accurate discrimination of playing techniques is completed through dynamic feature matching, significantly improving the accuracy and anti-interference stability of recognition in complex scenarios.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to an intelligent analysis method, system, device, and medium for musical instrument playing skills. Background Technology

[0002] The field of machine learning technology covers core aspects such as audio feature extraction, performance behavior recognition, model training, and classification prediction. This field collects audio data generated by musical instrument performances, extracts audio waveforms, spectral features, and temporal features, constructs data models, and completes the identification and analysis of performance-related information. Among them, the analysis of traditional musical instrument performance techniques refers to distinguishing and judging the playing actions and sound production methods such as fingering, bowing, and blowing techniques during the musical instrument performance process. The analysis of traditional musical instrument performance techniques usually relies on manual listening and annotation, combined with fixed audio feature calculations and threshold comparisons to complete the distinction and classification of performance techniques.

[0003] Current technologies rely on manual listening and annotation to process basic performance skills data. However, manual operation inherently involves subjective judgment biases, and different annotators struggle to maintain consistent standards for judging the same performance skill. The annotation process is time-consuming, inefficient, and poorly replicable, failing to meet the demands of batch processing large-scale performance audio data. Fixed audio feature calculation and threshold comparison processing modes are only suitable for ideal, noise-free laboratory environments. In live performance scenarios with stable background noise or sudden environmental noise, the noise signals directly interfere with the extraction and calculation of audio features, leading to systematic biases in the feature extraction results. Fixed thresholds cannot be dynamically adjusted according to changes in the environment, easily resulting in misjudgments and omissions of performance skills. For example, in live performance scenarios with venue echoes, audience conversations, and equipment background noise, the fixed threshold comparison mode cannot eliminate feature interference from various noises, misjudging spectral changes caused by noise as changes in performance skills. This ultimately leads to a significant decrease in the accuracy of performance skill analysis results, failing to meet the high-precision skill analysis requirements of complex live performance scenarios. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides an intelligent analysis method for musical instrument playing skills, comprising the following steps: S1: Collect pure audio of various standard playing techniques of the target instrument to form a standard instrument acoustic feature library. Simultaneously collect various typical environmental noises to construct a noise spectrum feature library. Merge the standard instrument acoustic feature library and the noise spectrum feature library to obtain a priori spectrum feature benchmark library. S2: Collect the live audio of the target instrument, call the prior spectrum feature benchmark library, generate the power spectrum of stationary noise in the live audio, perform constrained whitening filtering on the live audio, and obtain the preprocessed audio spectrum dataset; S3: Based on the preprocessed audio spectrum dataset, filter the signal frequency band that matches the standard frequency of the target instrument, remove the steady noise in the signal frequency band, compare the feature fit of the processed signal with the feature benchmark library of the prior spectrum feature, set the feature threshold, remove the sudden noise with the fit lower than the feature threshold, and obtain the pure performance spectrum feature matrix. S4: For the pure performance spectrum feature matrix, extract the time domain amplitude, frequency domain distribution and harmonic coherence data, perform three-dimensional joint weighted operation, and generate a set of technique feature quantification parameters; S5: Match the set of quantified skill features with the standard data in the prior spectrum feature benchmark library, calculate the feature matching degree, and generate the performance skill discrimination coefficient.

[0005] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Collect pure audio of various standard playing techniques of the target instrument, extract the corresponding spectral information of the audio, statistically analyze the spectral distribution characteristics, establish the correspondence between spectral characteristics and playing techniques, and generate a standard instrument acoustic feature library. S102: Collect various typical environmental noises, extract the corresponding spectrum information of the noises, record the correspondence between spectrum amplitude and frequency, and construct a noise spectrum feature library; S103: Call the standard musical instrument sound feature library and the noise spectrum feature library, merge the spectrum data in the two libraries, integrate the spectrum feature dimensions, unify the feature annotation rules, and fuse them to form a complete feature set, thus obtaining the prior spectrum feature benchmark library.

[0006] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Collect the live audio of the target instrument, call the prior spectrum feature benchmark library, match the live audio spectrum components with the stationary noise components in the prior spectrum feature benchmark library, locate the stationary noise distribution range in the live audio, and generate the stationary noise power spectrum. S202: Call the stable noise power spectrum, set the spectrum whitening constraint conditions, perform amplitude normalization processing on each spectrum component of the on-site audio, suppress the fluctuation amplitude of non-target signals in the spectrum components, and obtain the constrained whitening filter signal; S203: For constrained whitening filtered signals, extract the spectral amplitude and frequency distribution data at each time point, integrate the spectral data of multiple frames to form a time-series data set, and obtain a preprocessed audio spectrum dataset.

[0007] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the preprocessed audio spectrum dataset, retrieve the standard frequency data of the target instrument from the prior spectrum feature benchmark library, compare the frequency components of the spectrum dataset with the standard frequency values, and generate the matching frequency band of the target instrument. S302: For the target instrument matching frequency band, call the stable noise data in the prior spectrum feature reference library, remove the signal components in the frequency band that overlap with the stable noise spectrum, and obtain the frequency band purified performance signal; S303: Call the prior spectrum feature reference library, compare the frequency band purified performance signal with the spectrum features and amplitude features of the target instrument in the reference library, calculate the degree of fit between the two, set the feature threshold, remove signal components with a degree of fit lower than the feature threshold, and obtain the pure performance feature matrix.

[0008] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: For the pure performance spectrum feature matrix, extract the time domain amplitude data, frequency domain distribution data and harmonic coherence data within the matrix, integrate the three types of data to form a feature set, and generate a multi-dimensional performance feature set; S402: Based on a multi-dimensional performance feature set, a weighted superposition operation is performed on time-domain amplitude data, frequency-domain distribution data, and harmonic coherence data to unify the numerical dimensions of various data and fuse them to form a quantitative feature sequence, thereby obtaining a set of quantitative parameters for skill features.

[0009] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: For the skill feature quantization parameter set, extract the time domain and frequency domain related data, calculate the weight ratio of each parameter, and generate a parameter weight table; S502: Call the parameter weight table, combine it with the prior spectrum feature benchmark library, compare the parameter weights with the standard range, filter out the parameters that meet the requirements, and generate the parameter filtering results.

[0010] As a further aspect of the present invention, based on the parameter screening results, the matching degree of each parameter is calculated, the matching data is integrated and the values ​​are calibrated to obtain the performance skill discrimination coefficient.

[0011] A smart analysis system for musical instrument playing techniques, the system comprising: a feature benchmark construction module, a live signal preprocessing module, a noise precision removal module, a technique feature extraction module, and a playing technique discrimination module; The feature benchmark construction module is used to collect pure audio of various standard playing techniques of the target instrument to form a standard instrument sound feature library, and simultaneously collect various typical environmental noises to construct a noise spectrum feature library. The standard instrument sound feature library and the noise spectrum feature library are then fused to obtain the prior spectrum feature benchmark library. The on-site signal preprocessing module is used to acquire the on-site audio of the target instrument, call the prior spectrum feature benchmark library, generate the power spectrum of stationary noise in the on-site audio, perform constrained whitening filtering on the on-site audio, and obtain the preprocessed audio spectrum dataset. The noise removal module is used to filter signal frequency bands that match the standard frequency of the target instrument based on the preprocessed audio spectrum dataset, remove stable noise in the signal frequency band, compare the feature fit of the processed signal with the feature benchmark library of the prior spectrum feature, set the feature threshold, remove sudden noises with a fit lower than the feature threshold, and obtain a pure performance spectrum feature matrix. The skill feature extraction module is used to extract time-domain amplitude, frequency-domain distribution, and harmonic coherence data from the pure performance spectrum feature matrix, perform three-dimensional joint weighted calculation, and generate a set of skill feature quantification parameters. The performance skill discrimination module is used to match the set of skill feature quantification parameters with the standard data in the prior spectrum feature benchmark library, calculate the feature matching degree, and generate the performance skill discrimination coefficient.

[0012] A smart instrument playing technique analysis device includes: processor; Memory used to store the processor's executable instructions; The processor is used to execute any of the above-described intelligent analysis methods for musical instrument playing techniques.

[0013] A computer-readable storage medium storing a computer program for executing any one of the above-described intelligent analysis methods for musical instrument playing techniques. Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a priori spectral feature benchmark library is constructed by integrating standard performance audio with environmental noise. Stable and sudden noises are eliminated through constrained whitening filtering and hierarchical feature screening, while retaining all effective performance spectral features directly related to performance skills. The performance skills are refined by joint weighted quantization of multi-dimensional features in the time and frequency domains and harmonics. The performance skills are accurately identified through dynamic feature matching, which greatly improves the accuracy and anti-interference stability of recognition in complex scenarios. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0017] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides an intelligent analysis method for musical instrument playing skills, comprising the following steps: S1: Collect pure audio of various standard playing techniques of the target instrument to form a standard instrument acoustic feature library. Simultaneously collect various typical environmental noises to construct a noise spectrum feature library. Merge the standard instrument acoustic feature library and the noise spectrum feature library to obtain a priori spectrum feature benchmark library. The specific steps of S1 are as follows: S101: Collect pure audio of various standard playing techniques of the target instrument, extract the corresponding spectral information of the audio, statistically analyze the spectral distribution characteristics, establish the correspondence between spectral characteristics and playing techniques, and generate a standard instrument acoustic feature library. Based on different playing techniques of the target instrument in a standard performance environment, a closed acoustic space with no reverberation, no reflection, and no background noise was selected. A fixed pickup distance and a directional microphone were used for synchronous sound pickup. A uniform sampling rate and sampling precision were set, and each playing technique was repeatedly sampled multiple times. Audio segments with transient interference, amplitude overflow, and acquisition distortion were removed. The retained valid audio segments were processed from the time domain to the frequency domain. The audio waveform was converted into corresponding spectral data through discrete Fourier transform. The amplitude, phase, spectral peak position, bandwidth, and energy distribution information at different frequency points were extracted. Horizontal statistics were performed on the spectral data collected multiple times under the same playing technique to calculate the mean, variance, and dispersion of each frequency component, determining the stable spectral feature interval corresponding to the playing technique. The feature intervals corresponding to different playing techniques were independently encoded, establishing a one-to-one mapping relationship between spectral feature vectors and playing technique types. All mapping relationships were classified and stored according to the playing technique category, forming a structured and searchable standard instrument acoustic feature library. S102: Collect various typical environmental noises, extract the corresponding spectrum information of the noises, record the correspondence between spectrum amplitude and frequency, and construct a noise spectrum feature library; Typical indoor and outdoor environmental noise types were selected, and data collection was completed in a scenario where the noise source was stable and there were no transient strong interferences. The same sampling parameters as those used for the target musical instrument audio collection were adopted to ensure the consistency of the spectrum extraction dimensions. The collected noise audio was preprocessed to remove invalid data segments caused by touching or sudden sounds. The noise audio was converted into corresponding spectrum information through frequency domain analysis. A two-dimensional spectrum distribution was established with frequency as the horizontal axis and amplitude as the vertical axis. The frequency value and corresponding amplitude were recorded point by point to sort out the concentrated distribution of noise energy in different frequency bands, distinguish the amplitude variation patterns of low frequency, mid frequency, and high frequency bands, and perform weighted statistics on the data collected from multiple times for the same type of noise to determine the typical amplitude range of the noise at each frequency point. The noise was structured and labeled according to noise type, frequency distribution characteristics, and amplitude range to construct a noise spectrum feature library containing multiple types of environmental noise. S103: Call the standard musical instrument sound feature library and the noise spectrum feature library, merge the spectrum data in the two libraries, integrate the spectrum feature dimensions, unify the feature annotation rules, and fuse them to form a complete feature set, thus obtaining the prior spectrum feature benchmark library. The system retrieves spectral feature vectors corresponding to all playing techniques from the established standard instrument sound feature library and spectral data corresponding to various types of environmental noise from the noise spectral feature library. The two sets of data are aligned and matched at a unified frequency resolution. Instrument sound features and noise features at the same frequency points are merged and integrated. The number of dimensions, numerical types and storage formats of feature vectors are unified. The original feature annotations are standardized, and duplicate, conflicting and redundant annotations are deleted. A unified feature identification rule and indexing method are established. The merged spectral features are reorganized according to frequency dimension, feature type and source category to form a complete feature set containing the core features of instrument sound and noise interference features. A priori spectral feature benchmark library that can be used for subsequent processing is generated.

[0020] S2: Collect the live audio of the target instrument, call the prior spectrum feature benchmark library, generate the power spectrum of stationary noise in the live audio, perform constrained whitening filtering on the live audio, and obtain the preprocessed audio spectrum dataset; The specific steps of S2 are as follows: S201: Collect the live audio of the target instrument, call the prior spectrum feature benchmark library, match the live audio spectrum components with the stationary noise components in the prior spectrum feature benchmark library, locate the stationary noise distribution range in the live audio, and generate the stationary noise power spectrum. In actual performance scenarios, the on-site audio signal containing the sound of the target instrument and environmental interference is acquired through a sound pickup device. The on-site audio is processed by frame-by-frame windowing according to the sampling parameters consistent with the prior spectrum feature benchmark library. Each frame of time-domain signal is converted into frequency-domain signal to obtain the corresponding spectrum component. The stationary noise spectrum features and frequency distribution intervals stored in the prior spectrum feature benchmark library are read. The frequency points and amplitude information of each frame of the on-site audio are compared with the feature vectors of stationary noise in the benchmark library point by point. The feature similarity between the on-site spectrum component and the feature vectors of stationary noise in the benchmark library is calculated. A feature matching threshold is set to determine the component attribution. The spectrum component exceeding the matching threshold is determined to be a stationary noise related component. All frame signals are traversed and continuously distributed noise frequency intervals are marked. Power statistics are performed on the marked stationary noise frequency intervals. The power value at each frequency point is calculated. Each frequency and its corresponding power value are combined and arranged to form a complete stationary noise power distribution curve. S202: Call the stable noise power spectrum, set the spectrum whitening constraint conditions, perform amplitude normalization processing on each spectrum component of the on-site audio, suppress the fluctuation amplitude of non-target signals in the spectrum components, and obtain the constrained whitening filter signal; Read the stationary noise power spectrum, extract the power value and distribution range corresponding to each frequency point, set spectral whitening constraints based on the noise power distribution range and the effective frequency range of the target instrument, map the constraints to each frame of the spectral data of the live audio, traverse the amplitude information corresponding to each frequency point of the live audio, perform amplitude scaling on each frequency component with the stationary noise power spectrum as a reference, use normalization operation to map the amplitude of each frequency point to a unified value range, perform amplitude suppression operation on the spectral components in the frequency band corresponding to the stationary noise power spectrum, control the suppression amplitude and suppression range according to the constraints, preserve the amplitude characteristics of the frequency band where the target signal is located, complete amplitude adjustment and fluctuation suppression frame by frame, update the amplitude value of each frequency point point by point, combine the processed time domain and frequency domain signals to form the filtered output signal after constraint processing; S203: For constrained whitening filtered signals, extract the spectral amplitude and frequency distribution data at each time moment, integrate multi-frame spectral data to form a time-series data set, and obtain preprocessed audio spectral data; The output signal after constrained whitening filtering is subjected to time-series decomposition and inter-frame alignment. The frequency and amplitude values ​​corresponding to each moment of the signal are extracted frame by frame. The correspondence between each frequency point and amplitude point in each frame is recorded. The spectrum data of each frame is sorted in chronological order. A timestamp is added to each frame of spectrum data. The single-frame spectrum data is expanded into a time-series structure containing continuous information of multiple frames. The dimension and format of the spectrum data of each frame are unified. Redundant information and invalid data items between frames are removed. The spectral features of the time-series arrangement of multiple frames are integrated and encapsulated to construct a structured data set with time and frequency dimensions.

[0021] S3: Based on the preprocessed audio spectrum dataset, filter the signal frequency band that matches the standard frequency of the target instrument, remove the steady noise in the signal frequency band, compare the feature fit of the processed signal with the feature benchmark library of the prior spectrum feature, set the feature threshold, remove the sudden noise with the fit lower than the feature threshold, and obtain the pure performance spectrum feature matrix. The specific steps for S3 are as follows: S301: Based on the preprocessed audio spectrum dataset, retrieve the standard frequency data of the target instrument from the prior spectrum feature benchmark library, compare the frequency components of the spectrum dataset with the standard frequency values, and generate the matching frequency band of the target instrument. Read the frequency components and corresponding amplitude information of all frames in the dataset, sort the spectrum data of each frame in chronological order, and ensure that the data format is consistent with the prior spectrum feature benchmark library. Retrieve the standard frequency data corresponding to various standard playing techniques of the target instrument from the prior spectrum feature benchmark library, clarify the center frequency and allowable fluctuation range of each standard frequency, compare each frequency point in the preprocessed audio spectrum dataset with the standard frequency in the benchmark library point by point, calculate the deviation value between a single frequency point and the corresponding standard center frequency, set a frequency deviation threshold, which is set with reference to the pitch error range of the target instrument, and determine the frequency point with a deviation value less than or equal to the threshold as a matching frequency point. Classify and organize all matching frequency points, merge continuously distributed matching frequency points to form continuous frequency bands, perform boundary calibration on each continuous frequency band, remove isolated matching frequency points, mark the standard playing technique category corresponding to each frequency band, and integrate all calibrated continuous frequency bands to form the target instrument matching frequency band covering the main playing range of the target instrument. S302: For the target instrument matching frequency band, call the stable noise data in the prior spectrum feature reference library, remove the signal components in the frequency band that overlap with the stable noise spectrum, and obtain the frequency band purified performance signal; Spectral components and amplitude data are extracted one by one according to frequency band division, and a spectral distribution ledger for each frequency band is established. The spectral feature data of all stationary noises in the prior spectral feature benchmark library are retrieved, including the frequency distribution range, amplitude range and power characteristics of the noise. The spectral data of the matching frequency band of each target instrument is compared with the spectral data of the stationary noise band one by one. The similarity between the amplitude of each frequency point in the matching frequency band and the amplitude of the corresponding frequency point of the stationary noise is calculated. A similarity judgment threshold is set, which is set with reference to the amplitude fluctuation range of the stationary noise in the corresponding frequency band. Frequency components with similarity higher than the threshold are judged as components that overlap with the stationary noise spectrum. The components judged as overlapping are marked, and the marked overlapping components are deleted one by one. The remaining spectral components after deletion are subjected to amplitude completion processing to fill the amplitude gap caused by the deletion of noise components, ensuring the continuity of the spectrum within the frequency band. The processed spectral data of each frequency band are integrated to obtain the performance signal after frequency band purification. S303: Call the prior spectrum feature reference library, compare the frequency band purified performance signal with the spectrum features and amplitude features of the target instrument in the reference library, calculate the degree of fit between the two, set the feature threshold, remove signal components with a degree of fit lower than the feature threshold, and obtain the pure performance feature matrix. The spectral and amplitude features corresponding to various standard playing techniques of the target instrument are retrieved from the prior spectral feature benchmark library. Key parameters of the feature vectors in the benchmark library are extracted, including spectral peak position, amplitude peak value, spectral bandwidth, and energy distribution ratio. The spectral data of the purified performance signal in the frequency band is read, and the corresponding feature parameters such as spectral peak position and amplitude peak value are extracted frame by frame. The feature similarity calculation method is used to calculate the degree of fit between the purified signal features and each standard feature in the benchmark library frame by frame. The fit calculation adopts the weighted summation of feature parameters. The weight of each feature parameter is set according to its influence on the instrument sound recognition. After the calculation is completed, the fit values ​​of all frames are counted, the distribution law of fit is analyzed, and a reasonable range of fit is determined. Based on this range, a feature threshold is set. The feature threshold is the lower limit of the reasonable range of fit. The fit of all frames is judged one by one. Signal components with a fit value lower than the feature threshold are removed. These components are judged as signals corresponding to sudden noise. The remaining effective signal components are integrated in time and unified in dimension. They are arranged in matrix according to time order and frequency dimension to obtain the pure performance spectral feature matrix.

[0022] S4: For the pure performance spectrum feature matrix, extract the time domain amplitude, frequency domain distribution and harmonic coherence data, perform three-dimensional joint weighted operation, and generate a set of technique feature quantification parameters; The specific steps of S4 are as follows: S401: For the pure performance spectrum feature matrix, extract the time domain amplitude data, frequency domain distribution data and harmonic coherence data within the matrix, integrate the three types of data to form a feature set, and generate a multi-dimensional performance feature set; For the pure performance spectrum feature matrix, the matrix data is decomposed frame by frame in chronological order. The temporal amplitude data corresponding to each frame is extracted, including the amplitude peak, amplitude valley and amplitude change rate. By comparing frame by frame, the amplitude fluctuation parameters over time are calculated to ensure that the temporal data can fully reflect the strength changes of the performance signal. Then, the frequency domain distribution data is extracted to sort out the amplitude ratio, number of spectral peaks and distribution position of spectral peaks corresponding to each frequency point. Different frequency intervals are divided and the energy ratio of each interval is statistically analyzed to clarify the distribution law of frequency domain features. Subsequently, harmonic coherence data is extracted to identify the correspondence between the fundamental wave and each harmonic. The amplitude ratio, phase difference and coherence coefficient of the fundamental wave and harmonics are calculated to determine the degree of correlation between harmonics and fundamental wave. The extracted time domain, frequency domain and harmonic coherence data are aligned frame by frame to unify the data format and dimensions. Invalid items and redundant information in each type of data are removed. The three types of data are classified, labeled and integrated to form a set containing multi-dimensional features, generating a multi-dimensional performance feature set. S402: Based on a multi-dimensional performance feature set, a weighted superposition operation is performed on time-domain amplitude data, frequency-domain distribution data, and harmonic coherence data to unify the numerical dimensions of various data types, fuse them to form a quantitative feature sequence, and obtain a set of quantitative parameters for skill features; Dimensional analysis was performed on the three types of data to clarify the numerical range and dimensional differences of the time-domain amplitude, frequency-domain distribution, and harmonic coherence. A normalization method was used to map the three types of data to the same numerical range, eliminating the influence of dimensions on the calculation results. Weighting coefficients were then set for the three types of data, with the weights determined based on the degree of influence of each type of data on performance technique recognition. Frequency-domain distribution data had the highest weight, followed by harmonic coherence data, and time-domain amplitude data had the lowest weight. The discriminative power of each feature under different performance techniques was statistically analyzed to determine reasonable values ​​for each weight, ensuring that the weight allocation fits the actual performance scenario. Subsequently, a weighted superposition operation was performed, multiplying the normalized three types of data by their corresponding weights, and then summing the products to obtain the quantized feature value for each frame. The calculation was performed frame by frame, and the results were recorded. The quantized feature values ​​of all frames were sorted temporally, and quantized values ​​with abnormal fluctuations between frames were removed. Outliers were corrected, and all valid quantized feature values ​​were integrated to form a continuous quantized feature sequence, resulting in a set of quantized parameters for technique features.

[0023] S5: Match the set of quantitative parameters of skill features with the standard data in the prior spectrum feature benchmark library, calculate the feature matching degree, and generate the performance skill discrimination coefficient; The specific steps of S5 are as follows: S501: For the skill feature quantization parameter set, extract the time domain and frequency domain related data, calculate the weight ratio of each parameter, and generate a parameter weight table; For the set of quantitative parameters for skill features, the quantitative features related to time-domain amplitude change, frequency-domain energy distribution, and harmonic coherence are broken down. The distribution of each type of feature parameter in the time dimension is traversed, the dispersion and concentration interval of each parameter are statistically analyzed, the contribution ratio of a single type of parameter in the total number of features is calculated, the contribution ratio of each type of parameter is normalized so that all ratio values ​​meet the unified numerical constraints, the contribution ratio is adjusted according to the sensitivity of different parameters to the differentiation of performance skills, and the adjusted types of parameters and their corresponding ratio values ​​are arranged in a structured manner, classified and recorded in a fixed format to form a complete and orderly parameter weight table. S502: Call the parameter weight table, combine it with the prior spectrum feature benchmark library, compare the parameter weights with the standard range, filter out the parameters that meet the requirements, and generate the parameter filtering results; Read the values ​​and corresponding proportions of various parameters in the table, retrieve the parameter standard ranges and weight reference intervals corresponding to various standard performance techniques in the prior spectrum feature benchmark library, compare each parameter in the parameter weight table with the corresponding standard range in the benchmark library, determine whether the value of a single parameter is within the standard range, set the parameter compliance judgment criteria, which are set with reference to the characteristic stability interval of standard performance techniques, retain the parameter items that meet the judgment criteria, remove abnormal parameter items that exceed the standard range, classify and organize the retained valid parameters, rearrange them according to the original time sequence, and generate parameter filtering results; Based on the parameter screening results, the matching degree of each parameter is calculated, the matching data is integrated and the values ​​are calibrated to obtain the performance skill discrimination coefficient. Each parameter is compared with the corresponding standard parameter in the prior spectral feature benchmark library by difference calculation. The matching degree of each parameter is determined according to the difference. All matching degrees are weighted and summarized, and the summarization weight follows the corresponding proportion in the parameter weight table to obtain the overall matching value. The overall matching value is interval calibrated to make it fall into a uniform value range, eliminating the value offset caused by inter-frame fluctuations. The calibrated value is smoothed to remove local abrupt changes. The final value after processing is used as the discrimination criterion to obtain the performance skill discrimination coefficient.

[0024] A smart analysis system for musical instrument playing techniques, comprising: a feature benchmark construction module, a live signal preprocessing module, a noise precision removal module, a technique feature extraction module, and a playing technique discrimination module; The feature benchmark construction module is used to collect pure audio of various standard playing techniques of the target instrument to form a standard instrument sound feature library. Simultaneously, it collects various typical environmental noises to construct a noise spectrum feature library. The standard instrument sound feature library and the noise spectrum feature library are then merged to obtain the prior spectrum feature benchmark library. The on-site signal preprocessing module is used to acquire the on-site audio of the target instrument, call the prior spectrum feature benchmark library, generate the power spectrum of stationary noise in the on-site audio, perform constrained whitening filtering on the on-site audio, and obtain the preprocessed audio spectrum dataset. The noise removal module is used to filter signal frequency bands that match the standard frequency of the target instrument based on the preprocessed audio spectrum dataset, remove stable noise in the signal frequency band, compare the feature fit of the processed signal with the feature benchmark library of prior spectrum features, set feature thresholds, remove burst noise with a fit lower than the feature thresholds, and obtain a pure performance spectrum feature matrix. The technique feature extraction module is used to extract time-domain amplitude, frequency-domain distribution, and harmonic coherence data from the pure performance spectrum feature matrix, perform three-dimensional joint weighted calculations, and generate a set of technique feature quantification parameters. The performance technique discrimination module is used to match the set of quantitative parameters of technique features with the standard data in the prior spectrum feature benchmark library, calculate the feature matching degree, and generate the performance technique discrimination coefficient.

[0025] A smart instrument playing technique analysis device includes: processor; Memory used to store processor-executable instructions; The processor is used to execute any of the above-mentioned intelligent analysis methods for musical instrument playing techniques.

[0026] A computer-readable storage medium storing a computer program for executing any of the above-mentioned intelligent analysis methods for musical instrument playing techniques.

[0027] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A musical instrument playing skill intelligent analysis method, characterized in that, Includes the following steps: S1: Collect pure audio of various standard playing techniques of the target instrument to form a standard instrument acoustic feature library. Simultaneously collect various typical environmental noises to construct a noise spectrum feature library. Merge the standard instrument acoustic feature library and the noise spectrum feature library to obtain a priori spectrum feature benchmark library. S2: Collect the live audio of the target instrument, call the prior spectrum feature benchmark library, generate the power spectrum of stationary noise in the live audio, perform constrained whitening filtering on the live audio, and obtain the preprocessed audio spectrum dataset; S3: Based on the preprocessed audio spectrum dataset, filter the signal frequency band that matches the standard frequency of the target instrument, remove the steady noise in the signal frequency band, compare the feature fit of the processed signal with the feature benchmark library of the prior spectrum feature, set the feature threshold, remove the sudden noise with the fit lower than the feature threshold, and obtain the pure performance spectrum feature matrix. S4: For the pure performance spectrum feature matrix, extract the time domain amplitude, frequency domain distribution and harmonic coherence data, perform three-dimensional joint weighted operation, and generate a set of technique feature quantification parameters; S5: Match the set of quantified skill features with the standard data in the prior spectrum feature benchmark library, calculate the feature matching degree, and generate the performance skill discrimination coefficient.

2. The method of claim 1, wherein, The specific steps of S1 are as follows: S101: Collect pure audio of various standard playing techniques of the target instrument, extract the corresponding spectral information of the audio, statistically analyze the spectral distribution characteristics, establish the correspondence between spectral characteristics and playing techniques, and generate a standard instrument acoustic feature library. S102: Collect various typical environmental noises, extract the corresponding spectrum information of the noises, record the correspondence between spectrum amplitude and frequency, and construct a noise spectrum feature library; S103: Call the standard musical instrument sound feature library and the noise spectrum feature library, merge the spectrum data in the two libraries, integrate the spectrum feature dimensions, unify the feature annotation rules, and fuse them to form a complete feature set, thus obtaining the prior spectrum feature benchmark library.

3. The method of claim 2, wherein the method further comprises: The specific steps of S2 are as follows: S201: Collect the live audio of the target instrument, call the prior spectrum feature benchmark library, match the live audio spectrum components with the stationary noise components in the prior spectrum feature benchmark library, locate the stationary noise distribution range in the live audio, and generate the stationary noise power spectrum. S202: Call the stable noise power spectrum, set the spectrum whitening constraint conditions, perform amplitude normalization processing on each spectrum component of the on-site audio, suppress the fluctuation amplitude of non-target signals in the spectrum components, and obtain the constrained whitening filter signal; S203: For constrained whitening filtered signals, extract the spectral amplitude and frequency distribution data at each time point, integrate the spectral data of multiple frames to form a time-series data set, and obtain a preprocessed audio spectrum dataset.

4. The intelligent analysis method for musical instrument playing skills according to claim 3, characterized in that, The specific steps of S3 are as follows: S301: Based on the preprocessed audio spectrum dataset, retrieve the standard frequency data of the target instrument from the prior spectrum feature benchmark library, compare the frequency components of the spectrum dataset with the standard frequency values, and generate the matching frequency band of the target instrument. S302: For the target instrument matching frequency band, call the stable noise data in the prior spectrum feature reference library, remove the signal components in the frequency band that overlap with the stable noise spectrum, and obtain the frequency band purified performance signal; S303: Call the prior spectrum feature reference library, compare the frequency band purified performance signal with the spectrum features and amplitude features of the target instrument in the reference library, calculate the degree of fit between the two, set the feature threshold, remove signal components with a degree of fit lower than the feature threshold, and obtain the pure performance feature matrix.

5. The intelligent analysis method for musical instrument playing skills according to claim 4, characterized in that, The specific steps of S4 are as follows: S401: For the pure performance spectrum feature matrix, extract the time domain amplitude data, frequency domain distribution data and harmonic coherence data within the matrix, integrate the three types of data to form a feature set, and generate a multi-dimensional performance feature set; S402: Based on a multi-dimensional performance feature set, a weighted superposition operation is performed on time-domain amplitude data, frequency-domain distribution data, and harmonic coherence data to unify the numerical dimensions of various data and fuse them to form a quantitative feature sequence, thereby obtaining a set of quantitative parameters for skill features.

6. The intelligent analysis method for musical instrument playing techniques according to claim 5, characterized in that, The specific steps of S5 are as follows: S501: For the skill feature quantization parameter set, extract the time domain and frequency domain related data, calculate the weight ratio of each parameter, and generate a parameter weight table; S502: Call the parameter weight table, combine it with the prior spectrum feature benchmark library, compare the parameter weights with the standard range, filter out the parameters that meet the requirements, and generate the parameter filtering results.

7. The intelligent analysis method for musical instrument playing skills according to claim 6, characterized in that, Based on the parameter selection results, the matching degree of each parameter is calculated, the matching data is integrated and the values ​​are calibrated to obtain the performance skill discrimination coefficient.

8. An intelligent analysis system for musical instrument playing techniques, characterized in that, The system is used to implement the intelligent analysis method for musical instrument playing skills as described in any one of claims 1-7. The system includes: a feature benchmark construction module, a field signal preprocessing module, a noise precise removal module, a skill feature extraction module, and a playing skill discrimination module. The feature benchmark construction module is used to collect pure audio of various standard playing techniques of the target instrument to form a standard instrument sound feature library, and simultaneously collect various typical environmental noises to construct a noise spectrum feature library. The standard instrument sound feature library and the noise spectrum feature library are then fused to obtain the prior spectrum feature benchmark library. The on-site signal preprocessing module is used to acquire the on-site audio of the target instrument, call the prior spectrum feature benchmark library, generate the power spectrum of stationary noise in the on-site audio, perform constrained whitening filtering on the on-site audio, and obtain the preprocessed audio spectrum dataset. The noise removal module is used to filter signal frequency bands that match the standard frequency of the target instrument based on the preprocessed audio spectrum dataset, remove stable noise in the signal frequency band, compare the feature fit of the processed signal with the feature benchmark library of the prior spectrum feature, set the feature threshold, remove sudden noises with a fit lower than the feature threshold, and obtain a pure performance spectrum feature matrix. The skill feature extraction module is used to extract time-domain amplitude, frequency-domain distribution, and harmonic coherence data from the pure performance spectrum feature matrix, perform three-dimensional joint weighted calculation, and generate a set of skill feature quantification parameters. The performance skill discrimination module is used to match the set of skill feature quantification parameters with the standard data in the prior spectrum feature benchmark library, calculate the feature matching degree, and generate the performance skill discrimination coefficient.

9. An intelligent analysis device for musical instrument playing techniques, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is used to execute the intelligent analysis method for musical instrument playing skills as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which is used to execute the intelligent analysis method for musical instrument playing techniques as described in any one of claims 1-7.

Images