A method for automatically labeling structure of ethnic music based on rhythm and timbre features

Abstract

The application provides a kind of national music structure automatic marking method based on rhythm and timbre characteristics, comprising the following steps: S1) audio preprocessing and standardization;S2) rhythm cycle candidate extraction and stabilization;S3) timbre stability feature calculation and stable interval marking;S4) structure neighborhood construction and rhythm-timbre coupling weight calculation;S5) structure propagation update;S6) candidate boundary screening;S7) paragraph generation and rhythm cycle coverage constraint;S8) segment-level feature construction and hierarchical structure aggregation;S9) structure type labeling and archive output.The advantages of the application are: cross-work stability improvement, weak boundary detection capability enhancement, remote similar structure correlation, significant reduction of over-segmentation, and structured output of archive can be directly applied.

Description

Technical Field

[0001] This invention relates to the field of music information processing and audio structure analysis technology, specifically to an automatic annotation method for ethnic music structure based on rhythm and timbre features, which can be used for structural boundary detection, segment division, and structural type annotation of ethnic music audio. Background Technology

[0002] In the process of digital preservation of ethnic music, a large amount of audio data is stored in the form of complete recordings. This type of audio usually lacks clear segment division information and structural type identification, which leads to subsequent retrieval, analysis and management work relying on manual listening and identification, resulting in low efficiency.

[0003] Existing automatic music structure analysis methods mostly rely on energy changes, spectral differences, or fixed thresholds for segmentation. These methods are effective for music genres with clear rhythmic patterns and significant structural variations. However, folk music, in actual performance, often exhibits characteristics such as free rhythm, numerous ornaments, and gradual structural changes, making structural boundaries difficult to discern using a single acoustic indicator. Because existing methods generally employ preset fixed thresholds for boundary determination, these thresholds are ill-suited to the differences in dynamic range, performance style, and structural complexity between different works, resulting in high false positive and false negative rates and insufficient stability when applied across different works.

[0004] In addition, themes or motivations may appear repeatedly in different times and locations in some works, but existing methods usually only compare segments based on local feature similarity, lacking effective structural association and feature propagation and update mechanisms, making it difficult to identify these remotely similar structures, and resulting in low recall rates for repeated segments.

[0005] Meanwhile, existing methods often lack effective post-processing constraints after boundary detection, which can easily lead to excessively short segments and overly dense boundary divisions, resulting in over-segmentation and affecting the readability and practicality of structural analysis results.

[0006] Therefore, there is a need for a technical solution that can achieve stable structural division of ethnic music audio and output hierarchical structure annotation results with less reliance on human intervention. Summary of the Invention

[0007] The purpose of this invention is to provide an automatic annotation method for ethnic music structures based on rhythm and timbre features, which improves cross-work stability, enhances weak boundary detection capabilities, allows for association of remote similar structures, significantly reduces over-segmentation, and enables direct application of structured archive output.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: An automatic annotation method for the structure of ethnic music based on rhythm and timbre features includes the following steps: S1) Audio Preprocessing and Standardization Input ethnic music audio signal To ensure a uniform sampling rate and perform noise reduction processing, the frame signal is obtained through frame segmentation and windowing. , forming a frame set .

[0009] S1-1) Input and sampling rate are unified Audio signals of ethnic music (Continuous-time representation) converted into discrete signal When the input sampling rate is And the sampling rate is consistent with the system. When there is inconsistency, resampling is performed. A uniform sampling rate is used to ensure that rhythm period estimation has a consistent time scale.

[0010] S1-2) Amplitude Normalization and Optional Noise Reduction right Perform amplitude normalization to obtain When a high percentage of background noise is detected (which can be determined by energy statistics of silent segments), noise reduction can be performed to obtain... Otherwise, directly order = .

[0011] S1-3) Frame splitting and windowing Let the frame length be L (sampling points) and the frame shift be H. The frame signal is obtained by segmenting and windowing. , forming a frame set .

[0012] S1-4) Low-energy frame filtering (reduces false triggering of silence) Calculate energy per frame ,then At that time, the frame Frames marked as low-energy frames will not participate in subsequent structural neighborhood construction and boundary determination (but can be retained for final paragraph time localization).

[0013] The specific formula is as follows: (1) Frame-by-frame windowing ; Frame index, i=1,2,...,N; : Intra-frame sampling point index, m=0,1,...,L-1; : Preprocessed discrete audio signal; Frame shift (in pixels) determines the temporal resolution; L: Frame length (in pixels), which determines the frequency resolution; Window functions (such as the Hamming window) are used to reduce spectral leakage; : No. Windowed sample sequence of frames.

[0014] (2) Frame energy ; : No. Frame energy; Low energy threshold, which can be the global energy average. The ratio (e.g., 0.05) ~0.1 Alternatively, one can take a low quantile of the energy distribution.

[0015] S2) Candidate extraction and stabilization of rhythmic cycles For each frame, calculate the autocorrelation, take the delay corresponding to the autocorrelation peak as the candidate period, and then perform a moving average or median filter on the candidate period sequence to obtain the smoothed period. This reduces the jitter caused by free rhythm.

[0016] S2-1) Autocorrelation calculation Calculate the autocorrelation for each frame (or a short window consisting of several frames). ,Delay The search range is obtained by mapping the BPM range to avoid obtaining invalid cycles.

[0017] S2-2) Principal Period Candidate Extraction and Validity Determination The candidate period is taken from the delay corresponding to the relevant peak value. To prevent the accidental selection of harmonic or noise peaks, the following is required: (a) fall into (b) Peak value Greater than a relative threshold (e.g., greater than) (A certain rule-based judgment, which can be parameterized).

[0018] S2-3) Periodic sequence smoothing Perform moving average / median filtering on the candidate periodic sequences to obtain smoothed periods. This reduces the jitter caused by free rhythm.

[0019] The specific formula is as follows: (1) Autocorrelation ; : No. Frame in delay The autocorrelation value under the following conditions; Delay (point) ; : Delay boundary converted from the acceptable BPM range.

[0020] (2) Candidate period ; : No. Frame candidate rhythm period (point).

[0021] (3) Smoothing period ; : The rhythmic periodic characteristics after slippage; : Smoothing window radius (frames), configurable (e.g., 3-7).

[0022] S3) Calculation of timbre stability characteristics and marking of stability intervals Calculate the spectrum for each frame and spectral centroid Define the intensity of timbre variation ,when continuous Frames less than the threshold When this interval is reached, it is marked as a timbre-stable interval, which is used for subsequent boundary suppression and neighborhood construction.

[0023] S3-1) Spectrum Calculation Calculate the short-time Fourier transform for each frame to obtain Take the amplitude spectrum .

[0024] S3-2) Spectral centroid as a timbre indicator Calculate the spectral centroid Reflects the brightness / center of gravity of the tone. In folk music, changes in playing techniques and instrument switching often lead to significant changes in the center of gravity.

[0025] S3-3) Intensity of timbre variation and stable range Define the intensity of timbre variation .when continuous Frames less than the threshold When this interval is reached, it is marked as a timbre-stable interval, which is used for subsequent boundary suppression and neighborhood construction.

[0026] The specific formula is as follows: (1) STFT ; : The complex spectrum of the frame; Frequency index; Imaginary unit.

[0027] (2) Spectral centroid ; : No. Frame spectrum centroid (index unit, convertible to Hz); Amplitude spectrum.

[0028] (3) Intensity of timbre variation ; The intensity of timbre change between adjacent frames; Stable threshold, acceptable The lower quantiles or form of the distribution (system parameters); : Continuous frame count requirement (e.g.) ).

[0029] S4) Structural neighborhood construction and rhythm-timbre coupling weight calculation Build a candidate set for each frame It searches for similar frames only within a preset frame range; then it searches the candidate set. Frames that simultaneously satisfy both rhythmic and timbre similarity are selected to form a neighborhood. ; Calculate coupling distance And similarity is obtained through exponential mapping, and then in the neighborhood Internal normalization yields structural weights This makes the structural weights comparable and numerically stable.

[0030] S4-1) Candidate Set for Time Window (Controlling Complexity) For each frame Construct candidate set Only Search for similar frames within the specified range.

[0031] S4-2) Dual-threshold screening structure neighborhood Frames that simultaneously satisfy both rhythmic and timbre similarity are selected from the candidate set to form neighborhoods. : Similar rhythms: ; Similar timbre: .

[0032] S4-3) Coupling Distance and Normalized Weights Calculate coupling distance The similarity is obtained through exponential mapping, and then normalized within the neighborhood. This makes the weights comparable and the values ​​stable.

[0033] The specific formula is as follows: (1) Candidate set ; Candidate window radius (number of frames), parameterized configuration.

[0034] (2) Structural neighborhood ; : Rhythm cycle difference threshold (points); : Threshold for intensity difference in timbre change.

[0035] (3) Coupling distance ; , Scale parameter, used to balance the influence of two types of feature dimensions.

[0036] (4) Normalized weights ; : Structural weights, satisfying that the sum of weights in the neighborhood is 1, which facilitates stable subsequent propagation and convergence.

[0037] S5) Structural Propagation Update Define the basic frame structure feature vector , ,in, To smooth the period, For the spectral centroid, For frame energy; Preferably, additional dimensions such as spectral bandwidth and zero-crossing rate can also be added, as long as they are reproducible and interpretable indicators. However, it is recommended to keep the dimensions stable to facilitate implementation.

[0038] when At that time, according to structural weights Weighted aggregation and superposition of neighborhood features get ; when At that time, directly ordered .

[0039] The specific formula is as follows: ; : No. Original structural feature vector of the frame; : The updated structural feature vector; The structural weights obtained from S4; : Set of structural neighborhoods.

[0040] S6) Candidate Boundary Screening Calculate the change in feature differences when updating the frame structure feature vector. And perform adaptive thresholding. calculate; Frames satisfying the local maximum condition are used as candidate boundary points to avoid dense boundaries caused by continuous jitter. If the candidate boundary points satisfy... If it is, then it is confirmed as a boundary.

[0041] Furthermore, in step S6), for Calculate the mean and standard deviation Construct an adaptive threshold ,in, This is the threshold adjustment coefficient.

[0042] Further, in step S6), if the frame interval between two adjacent boundaries is less than... Then keep For larger boundary points, delete the smaller ones; among them, This is the minimum boundary interval, used to constrain boundary density.

[0043] The specific formula is as follows: (1) Change ; : Structural change; : L2 norm.

[0044] (2) Mean and standard deviation ; : Mean of change; : Standard deviation of change.

[0045] (3) Adaptive threshold ; Work-level adaptive threshold; Threshold adjustment coefficient (e.g., 0.6 to 1.2, configurable).

[0046] (4) Local maxima If the conditions are met, it will be included in the candidate set.

[0047] (5) Minimum interval Minimum boundary spacing (number of frames) is used to constrain boundary density and prevent over-density.

[0048] S7) Segment Generation and Rhythmic Cycle Coverage Constraints The entire audio segment is divided into an initial segment set based on the boundary set. ; If a certain segment If so, then the merge will be performed; where, For paragraph Frame length, The average rhythmic cycle of the work. This is the minimum coverage factor.

[0049] Furthermore, in step S7), if a certain segment When performing the merge, the segment-level similarity with its left and right segments is calculated, and the side with higher similarity is selected for merging to ensure structural continuity.

[0050] The specific formula is as follows: ; The average rhythmic cycle of the work; :paragraph Frame length; Minimum coverage factor (e.g., 1.2 to 2.0).

[0051] S8) Segment-level feature construction and hierarchical structure aggregation Update features on the intra-segment frame structure feature vector The average is used to obtain the segment-level features. Then calculate the distance between segments. ,in, Let be the distance between segments p and q; when And it satisfies the rhythm compatibility condition (e.g.) When ), the segment With Duan They aggregate into the same upper-level structural unit to form a hierarchical structure.

[0052] The specific formula is as follows: ; :part Segment-level feature vectors; Intersegment distance; Aggregation threshold; :part Inside The mean (segment period mean); Segment-level rhythm compatibility threshold.

[0053] S9) Structure Type Labeling and File Output Based on paragraph position, energy trend, rhythm stability, and the proportion of stable timbre intervals, regular annotation is performed, and a structured annotation file containing fields such as paragraph start and end time, paragraph number, structure type, paragraph-level feature index, and upper-level aggregation unit number is output. The file is also stored in the database for retrieval.

[0054] Further, in step S9), the system prefabricates or configures a set of structural tags, including but not limited to introduction, body, transition, and epilogue.

[0055] Specifically, the introduction is located at the beginning, with gradually increasing energy and a gradually stabilizing cycle; the main body has a stable cycle, a high proportion of stable timbre, and a relatively long duration; the transition has strong timbre changes, large periodic fluctuations, and obvious changes in the distance between sections; and the coda is located at the end, with an overall decrease in energy or a sparse rhythm.

[0056] Compared with the prior art, the present invention has the following advantages: (1) Improved stability across works: Since the boundary threshold is determined based on the structural change statistics within the current work, the problem of false detection and false negative detection caused by fixed threshold in different works can be reduced.

[0057] (2) Enhanced weak boundary detection capability: For structural boundaries with insignificant energy fluctuations but changes in rhythm or timbre, the present invention can still achieve identification through joint analysis of rhythm features and timbre features.

[0058] (3) Remote similar structures can be associated: For thematic paragraphs that recur at different times and locations, the present invention can enhance the consistency of their features through structural propagation updates, thereby improving the recognition effect of recurring structures.

[0059] (4) Significantly reduce oversegmentation: By combining minimum boundary interval constraints, rhythmic period coverage constraints and adjacent segment merging strategies, the occurrence of short segments and dense boundaries can be reduced, thereby reducing oversegmentation.

[0060] (5) The structured output of the archives can be directly applied: This invention can output the start and end time of paragraphs, paragraph types and hierarchical aggregation results, which facilitates the storage, retrieval and subsequent display of ethnic music archives. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0062] Figure 2 This is a graph showing the rhythmic cycle change in step S2 of Embodiment 1 of the present invention.

[0063] Figure 3 This is a curve showing the change in timbre stability in step S3 of embodiment 1 of the present invention.

[0064] Figure 4 This is the structural neighborhood relationship heatmap in step S4 of embodiment 1 of the present invention.

[0065] Figure 5 This is a comparison diagram of the structural features before and after propagation in step S5) of Embodiment 1 of the present invention.

[0066] Figure 6 This is a diagram showing the structural boundary detection results in step S6) of Embodiment 1 of the present invention.

[0067] Figure 7 This is a diagram showing the segmentation result in step S7 of embodiment 1 of the present invention.

[0068] Figure 8 This is a diagram showing the hierarchical structure aggregation result in step S8 of Embodiment 1 of the present invention. Detailed Implementation

[0069] The embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Example 1

[0070] This embodiment uses a video file of the ethnic instrumental ensemble piece "Fisherman's Triumphant Song" as the input sample, and extracts the audio signal from the video file as the object to be processed. The total duration of the audio is 366.968 seconds, the sampling rate is 22050Hz, the number of channels is 1, and the file size is 21.836 MB. In this embodiment, the system uses a frame length of 2048 points and a frame shift of 512 points for short-time analysis, forming a total of 15801 analysis frames. The sample has the rhythmic progression, timbre changes, and segmental repetition features common in ethnic instrumental works, which can be used to verify the applicability of this invention for automatically annotating the structure of ethnic music.

[0071] S1) Audio Preprocessing and Standardization The system extracts audio from the input video file and converts the audio into a mono signal. Subsequent processing includes sampling rate unification, amplitude normalization, and frame-by-frame windowing. For low-energy frames, the system retains their temporal index for final segment localization, but reduces its involvement in structural neighborhood construction and boundary determination to minimize false triggering caused by silent or weak-energy segments. After processing, the input audio is converted into a frame sequence suitable for short-term analysis.

[0072] S2) Candidate extraction and stabilization of rhythmic cycles After framing, the autocorrelation function is calculated within a short-time analysis window, and the delay corresponding to the autocorrelation peak is selected as a candidate rhythmic period within a preset rhythmic period range. To reduce the influence of free beats, ornaments, and local performance fluctuations, median filtering is applied to the candidate period sequence to obtain a smoothed rhythmic period sequence.

[0073] like Figure 2 As shown, the rhythmic cycle of the work is not completely constant across different time intervals, but exhibits slow fluctuations and local adjustments. This indicates that using only a fixed beat or a fixed time window is insufficient to accurately reflect the structural changes in traditional Chinese instrumental music. This invention, through rhythmic cycle estimation and smoothing, can obtain a more stable rhythmic description, providing a foundation for subsequent structural boundary detection.

[0074] S3) Calculation of timbre stability characteristics and marking of stability intervals After completing the rhythm period extraction, a short-time Fourier transform is performed on each frame to calculate the spectral centroid, and the intensity of timbre change is obtained based on the change of the spectral centroid between adjacent frames. When the intensity of timbre change remains at a low level in consecutive frames, the corresponding interval is recorded as the timbre stable interval.

[0075] like Figure 3 As shown, some intervals exhibit high timbre stability, indicating a relatively continuous timbre pattern within the corresponding segment. Conversely, locations where the intensity of timbre changes increases often correspond to changes in performance technique, orchestration, or phrase transitions. Therefore, timbre stability, along with rhythmic periodicity, can be used to suppress missegmentation within stable segments and enhance the ability to identify weak boundary locations.

[0076] S4) Structural neighborhood construction and rhythm-timbre coupling weight calculation In this stage, the system uses the smooth rhythm period Ti and the timbre variation intensity Si as a basis to search for candidate frames that simultaneously satisfy the conditions of rhythmic similarity and timbre similarity with the current frame within a preset time range, and constructs structural neighborhood relationships accordingly. For frames that enter the structural neighborhood, the coupling distance is calculated based on the rhythmic difference and timbre difference, and then the structural weights are obtained through exponential mapping and normalization.

[0077] like Figure 4As shown in the figure, the brighter areas indicate a high structural similarity between corresponding frames. Besides the local continuity near the main diagonal, similarities also exist between some non-adjacent time positions, indicating the presence of repetitive or similar rhythmic-timbre structures in the work. The structural weights obtained in this stage are used for subsequent structural propagation updates.

[0078] S5) Structural Propagation Update The system uses indicators such as smooth rhythm period, spectral centroid and frame energy to form basic frame structural features hi; when there is a structural neighborhood in the current frame, the neighborhood features are weighted and converged according to the structural weights obtained in S4), and fused with the basic features of the current frame to obtain the propagated updated structural features hi'; when there is no effective structural neighborhood in the current frame, its basic structural features are directly retained.

[0079] like Figure 5 The figure shows the changes in features before and after structural propagation. The feature curves after propagation are more stable than before propagation, while retaining the positions of the main structural changes. The results indicate that structural propagation can enhance the feature consistency between similar segments and reduce the interference of local noise or instantaneous performance changes on structural segmentation.

[0080] S6) Candidate Boundary Screening After obtaining the updated structural features, the structural changes between adjacent frames are calculated, and an adaptive threshold is constructed based on the mean and standard deviation within the current work. Locations that satisfy the local maximum condition and exceed the adaptive threshold are identified as candidate boundaries; for candidate boundaries that are too close, only boundary points with large structural changes are retained.

[0081] like Figure 6 As shown, the detected structural boundary positions are marked on the original waveform. It can be seen that the boundaries do not appear at equal intervals, but are distributed according to the rhythmic progression and timbre changes within the piece, indicating that the boundary detection of this invention does not rely on fixed-duration segmentation. The detection yielded a total of 14 structural boundary points in this embodiment, with the first valid boundary located at approximately 21.269 seconds and the last valid boundary located at approximately 352.943 seconds.

[0082] S7) Segment Generation and Rhythmic Cycle Coverage Constraints Based on the aforementioned boundary set, the system divides the entire audio segment into 15 structural segments and merges excessively short segments by combining segment length and rhythmic period coverage constraints.

[0083] like Figure 7 As shown, the segmented sections vary in length, with the longest segment being approximately 42.075 seconds, the shortest approximately 14.025 seconds, and the average segment length approximately 24.465 seconds. This segmentation reflects the structural connections between different musical phrases or sections in the example audio.

[0084] S8) Segment-level feature construction and hierarchical structure aggregation Based on the segmentation, the system averages the updated frame structure features propagated within each segment to obtain segment-level feature vectors; then it calculates the inter-segment distance and, in conjunction with the rhythm compatibility condition, aggregates similar segments into the same upper-level structural unit.

[0085] like Figure 8 The diagram shows the hierarchical structure aggregation result of this embodiment. After aggregation, 15 bottom-level structural segments are merged into 4 upper-level structural units. This result reflects the correspondence between local segments and the overall structure of the work, facilitating subsequent segment retrieval, structural display, and musicological analysis within the digital archive of ethnic music.

[0086] S9) Structure Type Labeling and File Output The system performs regular structural type labeling on paragraphs based on paragraph position, energy trend, rhythm stability, and the proportion of stable timbre intervals, and outputs structured labeling results including paragraph start and end times, paragraph number, structural type, and upper-level structural unit number.

[0087] Table 1 shows the statistical results of the structure types in this embodiment.

[0088] Table 1 Statistical Results of Structure Types Introduction 1 main body 10 transition 3 end 1 As shown in Table 1, this embodiment ultimately yields one introduction segment, ten main body segments, three transition segments, and one epilogue segment. The above results demonstrate that this invention not only provides boundary locations but also further generates segment annotations with structural semantics.

[0089] In summary, this embodiment, using "Fisherman's Triumphant Song" as an example, obtained structured annotation results after processing, including the start and end times of each segment, structural type, and upper-level structural unit number. This demonstrates that the present invention can automatically annotate the structure of ethnic music audio without relying on manual segment-by-segment listening, and can provide a data foundation for ethnic music archive management and subsequent retrieval and utilization.

[0090] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the concept of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An automatic annotation method for the structure of ethnic music based on rhythm and timbre features, characterized in that... Includes the following steps: S1) Audio Preprocessing and Standardization Input ethnic music audio signal To ensure a uniform sampling rate and perform noise reduction processing, the frame signal is obtained through frame segmentation and windowing. , forming a frame set ; S2) Candidate extraction and stabilization of rhythmic cycles For each frame, calculate the autocorrelation, take the delay corresponding to the autocorrelation peak as the candidate period, and then perform a moving average or median filter on the candidate period sequence to obtain the smoothed period. This reduces the jitter caused by free rhythm; S3) Calculation of timbre stability characteristics and marking of stability intervals Calculate the spectrum for each frame and spectral centroid Define the intensity of timbre variation ,when continuous Frames less than the threshold At this time, the interval is marked as the timbre stable interval, which is used for subsequent boundary suppression and neighborhood construction; S4) Structural neighborhood construction and rhythm-timbre coupling weight calculation Build a candidate set for each frame It searches for similar frames only within a preset frame range; then it searches the candidate set. Frames that simultaneously satisfy both rhythmic and timbre similarity are selected to form a neighborhood. ; Calculate coupling distance And similarity is obtained through exponential mapping, and then in the neighborhood Internal normalization yields structural weights This makes the structural weights comparable and numerically stable; S5) Structural Propagation Update Define the basic frame structure feature vector , ,in, To smooth the period, For the spectral centroid, For frame energy; when At that time, according to structural weights Weighted aggregation and superposition of neighborhood features get ; when At that time, directly ordered ; S6) Candidate Boundary Screening Calculate the change in feature differences when updating the frame structure feature vector. And perform adaptive thresholding. calculate; Frames satisfying the local maximum condition are selected as candidate boundary points. If the candidate boundary points satisfy... If so, then it is confirmed as a boundary; S7) Segment Generation and Rhythmic Cycle Coverage Constraints The entire audio segment is divided into an initial segment set based on the boundary set. ; If a certain segment If so, then the merge will be performed; where, For paragraph Frame length, The average rhythmic cycle of the work. Minimum coverage factor; S8) Segment-level feature construction and hierarchical structure aggregation Update features on the intra-segment frame structure feature vector The average is used to obtain the segment-level features. Then calculate the distance between segments. ,in, Let be the distance between segments p and q; when And when the rhythm compatibility condition is met, the segment With Duan They aggregate into the same upper-level structural unit to form a hierarchical structure; S9) Structure Type Labeling and File Output Based on paragraph position, energy trend, rhythm stability, and the proportion of stable timbre intervals, regular annotation is performed, and a structured annotation file containing fields such as paragraph start and end time, paragraph number, structure type, paragraph-level feature index, and upper-level aggregation unit number is output. The file is also stored in the database for retrieval.

2. The method for automatically annotating the structure of ethnic music based on rhythm and timbre features according to claim 1, characterized in that: In step S6), for Calculate the mean and standard deviation Construct an adaptive threshold ,in, This is the threshold adjustment coefficient.

3. The method for automatically annotating the structure of ethnic music based on rhythm and timbre features according to claim 1 or 2, characterized in that: In step S6), if the frame interval between two adjacent boundaries is less than... Then keep For larger boundary points, delete the smaller ones; among them, This is the minimum boundary interval, used to constrain boundary density.

4. The method for automatically annotating the structure of ethnic music based on rhythm and timbre features according to claim 1, characterized in that: In step S7), if a certain segment When performing the merge, the segment-level similarity with its left and right segments is calculated, and the side with higher similarity is selected for merging to ensure structural continuity.

5. The method for automatically annotating the structure of ethnic music based on rhythm and timbre features according to claim 1, characterized in that: In step S9), the system prefabricates or configures a set of structural tags, including but not limited to introduction, body, transition, and epilogue.

Images