High-quality bird sound rapid screening method based on acoustic index fusion judgment
By combining the DCI, ADI, and FADI indices into a joint decision method, the accuracy and stability issues of bird sound screening in complex soundscapes are solved, achieving efficient and concise bird sound segment screening, which is suitable for rapid screening of large-scale bird sound recording data and ecological environment monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-13
- Publication Date
- 2026-05-15
AI Technical Summary
In complex acoustic environments, existing data filtering methods based on acoustic indices are easily affected by background noise, leading to noise segments being misjudged as valid data or real bird call segments being incorrectly rejected, thus reducing the accuracy and completeness of the filtering results.
An acoustic index-based fusion decision method is adopted. By calculating indices such as DCI, ADI, and FADI and combining them with frequency band energy distribution characteristics, adaptive discrimination conditions are constructed to jointly decide on audio data segments and output high-quality bird sound segments.
It improves the accuracy and stability of screening under complex background noise conditions, simplifies the algorithm structure, improves data utilization quality and processing efficiency, and is suitable for rapid screening of large-scale bird sound recording data.
Smart Images
Figure CN122050434A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rapid assessment of bird diversity, and specifically to a rapid screening method for high-quality bird calls based on acoustic index fusion decision. Background Technology
[0002] Acoustic indices were initially proposed to quantify the structural characteristics of soundscapes from a statistical perspective, and further reflect the level of bioacoustic activity and biodiversity status in local areas. In bird diversity monitoring applications, indices such as the Acoustic Diversity Index (ADI) and the Acoustic Complexity Index (ACI) have been widely used to estimate the overall characteristics of bird calls in terms of spectral distribution, temporal and frequency intensity variations, etc. Therefore, acoustic indices can serve as core quantitative indicators for diversity assessment and also have the potential to enable rapid screening of high-quality segments.
[0003] In long-term, continuous bird call monitoring scenarios, automated acquisition systems typically generate massive amounts of raw recording data, of which a relatively small proportion contain valid bird call segments. Calculating the acoustic index directly from all the data is not only computationally expensive, but noise-dominated data significantly reduces the stability and ecological interpretability of the index results. Therefore, effectively screening the raw recordings before calculating the acoustic index, retaining audio segments with a high proportion of bird calls and a high signal-to-noise ratio (SNR), is a crucial prerequisite for improving the efficiency and reliability of acoustic index applications.
[0004] It should be noted that the requirements for data selection are not entirely consistent across different application objectives. For automatic bird identification tasks, emphasis is typically placed on bird call segments with a single sound source, clear structure, and high SNR to avoid interference from multiple sound sources on classification performance. However, in acoustic index applications aimed at assessing bird diversity, the focus is more on the overall spectral distribution and activity intensity characteristics of bird calls within the analysis period. The requirement for the purity of single sound sources is relatively lower, but the bird call signal must dominate the time-frequency structure, and the interference of background noise on statistical features should be minimized. Given the robust calculation of the bird diversity acoustic index and its practical application requirements, the design of subsequent data selection methods should primarily aim to meet the audio data quality requirements for acoustic index calculation. Existing data selection methods based on acoustic indices often rely on a single indicator or empirical threshold for judgment, such as directly selecting segments with high ACI or ADI as candidate data. However, in complex soundscape environments, this type of method often faces two prominent problems: on the one hand, background noise (especially man-made noise and broadband random noise) may significantly affect the calculation of the exponent value under low SNR conditions, causing a large number of noise segments to be misjudged as valid bird sound data; on the other hand, when the bird sound intensity is weak or partially masked by noise, real bird sound segments may be incorrectly removed due to low exponent values, thereby reducing the integrity of the screening results. Summary of the Invention
[0005] The purpose of this invention is to provide a method for rapid screening of high-quality bird calls based on acoustic index fusion decision.
[0006] The technical solution to achieve the purpose of this invention is as follows: a method for rapid screening of high-quality bird calls based on acoustic index fusion decision, comprising the following steps:
[0007] Step 1: Load and preprocess the acquired audio data to obtain audio data segments for analysis;
[0008] Step 2: For each data segment in Step 1, estimate the average background noise power of the data segment; based on the estimated noise power, calculate the noise suppression complexity index which has a positive correlation with bird call activity, and complete the preliminary screening of candidate audio data segments accordingly.
[0009] Step 3: For the candidate audio data segments obtained in Step 2, calculate two types of acoustic diversity indices to describe the frequency band energy distribution characteristics, and further filter the candidate audio data segments based on the frequency band activity characteristics.
[0010] Step 4: Based on the acoustic diversity index and its corresponding frequency band features, construct adaptive discrimination conditions, perform joint judgment on audio data segments, and output high-quality bird sound segments.
[0011] A high-quality bird call rapid screening system based on acoustic index fusion decision, used to implement the above method, includes:
[0012] The first module is used to load and preprocess the acquired audio data to obtain audio data segments for analysis.
[0013] The second module estimates the average background noise power of each data segment obtained in the first module; based on the estimated noise power, it calculates the noise suppression complexity index, which has a positive correlation with bird call activity, and completes the preliminary screening of candidate audio data segments accordingly.
[0014] The third module calculates two types of acoustic diversity indices to describe the frequency band energy distribution characteristics for the candidate audio data segments obtained in the second module, and further filters the candidate audio data segments based on the frequency band activity characteristics.
[0015] The fourth module constructs adaptive discrimination conditions based on the acoustic diversity index and its corresponding frequency band characteristics, performs joint judgment on audio data segments, and outputs high-quality bird sound segments.
[0016] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.
[0017] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the above-described method.
[0018] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.
[0019] Compared with existing technologies, the beneficial effects of this invention are as follows: 1) This invention uses a combination of multiple acoustic indices (such as DCI, ADI, and FADI) for joint decision-making, which, compared with traditional methods based on a single acoustic feature or static threshold, can more effectively cope with complex background noise and varied natural soundscapes, improving the accuracy and stability of high-quality bird sound clip selection; 2) This invention adopts an index calculation and decision-making process with clear rules and few parameters, without involving complex model training or repeated parameter adjustments. The algorithm structure is simple and highly interpretable, making it easy to deploy and maintain in actual engineering systems, which is conducive to improving the repeatability and consistency of acoustic index selection results; 3) Due to the low computational complexity of the method of this invention, it can quickly perform offline screening of large-scale bird sound recording data, significantly improving overall processing efficiency and data utilization quality.
[0020] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description
[0021] Figure 1 This is a flowchart of a rapid screening method for high-quality bird calls based on acoustic index fusion decision.
[0022] Figure 2 This is a histogram distribution of DAI values for all segments and the top 10% of segments by DAI ranking.
[0023] Figure 3 Compare the candidate fragments with the final selected fragments using histograms of ADI (top) and FADI (bottom).
[0024] Figure 4 This document presents the visualization results of bird call screening for 1201 data segments using a high-quality bird call screening method based on acoustic index fusion decision. Detailed Implementation
[0025] This invention proposes a rapid screening method for high-quality bird calls based on acoustic index fusion decision. The method includes: first, automatically loading continuously acquired audio files within a preset date range, parsing and sorting the acquisition time information, resampling and unifying the channels of the audio data, segmenting the audio signal by natural minutes and completing time labeling. Then, estimating the background noise power for each audio segment and calculating the Denoising Acoustic Complexity Index (DACI), sorting the segments according to the DCI values, and selecting audio segments within a preset proportion range as a candidate segment set. Based on this, calculating the Acoustic Diversity Index (ADI) and the Frequency-dependent Acoustic Diversity Index (FADI) based on frequency-varying threshold detection for the candidate segments, sorting and cumulatively analyzing their frequency band energy distribution, and extracting the number of effective frequency bands. Further, constructing a dynamic discrimination threshold based on the number of effective frequency bands, and obtaining the final high-quality bird call segment screening result through multi-condition joint decision using ADI and FADI. This invention has the advantages of high computational efficiency and strong robustness to noise interference. It is suitable for the automated analysis and processing of large-scale field data collection, and can effectively improve the quality and usability of front-end acoustic data. It has important application value for bird diversity assessment in ecological environment monitoring.
[0026] Combination Figure 1 A rapid method for screening high-quality bird calls based on acoustic index fusion decision is proposed, with the following steps:
[0027] Step 1: Load and preprocess the acquired data segments:
[0028] Step 1-1: Within a preset date range, automatically load continuously acquired WAV audio files; based on the standard file naming rules adopted by the acquisition device, parse and extract the acquisition timestamp information corresponding to each audio file;
[0029] Step 1-2: Sort the loaded audio files according to their chronological order based on the timestamp information obtained from the parsing.
[0030] Steps 1-3: Perform uniform resampling on the sorted audio files and extract single-channel audio signals from the multi-channel audio to ensure consistency in subsequent processing. In a specific example of this invention, the sampling rate is set to 48 kHz;
[0031] Steps 1-4: Divide the processed audio signal into multiple consecutive audio segments according to natural minutes, with each segment having a fixed duration of 1 minute; simultaneously, record the actual acquisition time corresponding to each audio segment. In this specific embodiment of the invention, a total of 1201 audio segments were obtained.
[0032] Step 2: For each data segment in Step 1, estimate the background noise power of the data segment and calculate the DCI based on it. The DCI is improved by introducing a floating threshold mechanism. The DCI values of all audio segments are sorted, and audio segments with DCI values ranking within a preset proportion are selected as the candidate bird sound segment set; the remaining segments are discarded.
[0033] Step 2-1: For each audio segment obtained in Step 1, estimate its background noise power using the histogram noise estimation method for subsequent acoustic index calculation. The specific implementation is as follows: First, the data is processed by frame segmentation. Each frame is windowed and then subjected to a short-time discrete Fourier transform (SFT) to obtain the time-frequency power spectrum P(k,q), 1≤k≤K, 1≤q≤Q, where k and q are the frame number and frequency point number, respectively, K is the total number of frames in the analysis period, and Q is the total number of frequency points within the analysis bandwidth. In this specific example, the frame length and frame shift are both 10 ms, and the number of samples for the SFT is 512. A Hamming window is applied before the SFT. Second, the power of all frames corresponding to each frequency point is statistically analyzed. The first peak value is found in the probability histogram, and the corresponding power is the narrowband noise power of that frequency point.
[0034] Step 2-2: Based on the narrowband noise power estimation results, calculate the DCI. The DCI improves the stability of the ACI calculation by normalizing the energy of different frequency bands and introducing a floating threshold mechanism. A larger value indicates a greater number of bird calls and a higher level of activity. The calculation method and specific process of the DCI index are described in Chinese Invention Patent Publication No. CN119170049B, and will not be repeated here. In this specific example of the invention, the lowest frequency for DCI calculation is set to 0.5 kHz, and the highest frequency is set to 12 kHz.
[0035] Steps 2-3: Sort the DCI values corresponding to all audio segments to obtain a sequence of audio segments arranged from high to low acoustic complexity;
[0036] Steps 2-4: Select audio segments with DCI values within a preset range from the sorting results as a candidate segment set. In a specific example of this invention, the top 10% of audio segments are selected.
[0037] Step 3: For the candidate segment set obtained in Step 2, calculate ADI and FADI respectively. ADI describes the energy distribution of an audio segment across different frequency bands; FADI introduces an adaptive energy threshold based on ADI to suppress low-energy or noise-dominated frequency bands, improving discrimination stability in strong interference environments. Further, sort the frequency band energy sequences corresponding to ADI and FADI respectively, and calculate the cumulative energy; when the cumulative energy reaches a preset proportion of the total energy, determine the corresponding effective frequency band number (strong_bands_ADI and strong_bands_FADI), and determine whether the preset frequency band number condition has been met. Segments that do not meet the requirements are removed.
[0038] Step 3-1: For the candidate segment set obtained in Step 2, calculate the ADI value for each audio segment. ADI quantifies the frequency distribution range in a soundscape by statistically analyzing the energy distribution of the recorded signal across different frequency bands. The core idea of ADI is that the more species and the richer the sound sources, the wider the range of sound energy distribution on the frequency axis. ADI divides the entire frequency range into several frequency bands and calculates Shannon entropy based on the energy of each frequency band to reflect the diversity of frequency distribution. A higher value usually indicates significant acoustic activity in multiple frequency bands, suggesting a richer variety and higher activity of species calls during that time period. The calculation method and specific process of the ADI index are existing technologies and will not be elaborated here in this embodiment. In the specific example of this invention, the lowest frequency for ADI calculation is set to 0 kHz, and the highest frequency is set to 24 kHz.
[0039] Step 3-2: For the candidate segment set obtained in Step 2, calculate the FADI value for each audio segment. FADI is a bird acoustic diversity index method that is less sensitive to noise, addressing the fundamental shortcomings of ADI. The time-spectrum binarization processing of FADI uses narrowband floating detection thresholds set based on the average noise power at each frequency point while meeting certain SNR requirements. The full-scale relative level (dB Full Scale, dBFS) threshold of ADI is used as the lower limit of the threshold at each frequency point. The other index calculation processes are the same as ADI, and the specific process is described in Chinese Invention Patent Publication No. CN114913869A, which will not be repeated here. In this specific example of the invention, the lowest frequency for FADI calculation is set to 0 kHz, the highest frequency to 24 kHz, and the lowest signal-to-noise ratio threshold to 13 dB.
[0040] Step 3-3: Sort the energy values of each frequency band from high to low for the frequency band energy sequences corresponding to ADI and FADI respectively;
[0041] Steps 3-4: Calculate the cumulative energy of the sorted frequency band energy sequence; when the cumulative energy reaches a preset proportion of the total energy, determine the required number of frequency bands, denoted as strong_bands_ADI and strong_bands_FADI respectively. In a specific example of this invention, the preset proportion is set to 90%.
[0042] Steps 3-5: Determine whether strong_bands_ADI and strong_bands_FADI meet the preset frequency band quantity condition. If not, discard the segment. In a specific embodiment of the present invention, the preset frequency band quantity is set to 6.
[0043] Step 4: For the set of filtered segments obtained in Step 3, construct dynamic discrimination thresholds using strong_bands_ADI and strong_bands_FADI, and introduce an empirical adjustment coefficient to adaptively correct the thresholds. When the ADI and FADI values of an audio segment simultaneously exceed the corresponding dynamic thresholds, the audio segment is judged as a high-quality bird sound segment and output as the final filtering result.
[0044] Step 4-1: Using strong_bands_ADI and strong_bands_FADI as indicators of frequency band activity, introduce empirical adjustment coefficients. Dynamic discrimination thresholds T are constructed for ADI and FADI respectively. ADI T FADI :
[0045]
[0046]
[0047] In specific examples of the present invention Set to 0.618;
[0048] Step 4-2: Determine whether the ADI value and FADI value of the audio segment exceed the corresponding dynamic threshold at the same time. When the above judgment conditions are met at the same time, the corresponding audio segment is judged as a high-quality bird sound segment and output as the final filtering result.
[0049] Figure 2 The histogram distribution of DACI values for all segments and the top 10% selected segments is shown. It can be seen that the range with the highest DACI values for all segments is in the low-value region, indicating a large number of invalid segments in the original data. The DACI values for the top 10% of segments are distributed in the range of 3700-4700, indicating that the selected data segments should contain a large amount of time-varying sound rather than background noise. This demonstrates that DACI is a good initial screening indicator for quickly removing invalid segments.
[0050] Figure 3 The comparison of ADI and FADI histograms for candidate and final selected segments clearly shows that the segments that ultimately pass the selection exhibit a more concentrated distribution of high values in both ADI and FADI. In other words, DDI is primarily used to eliminate a large amount of invalid noise, while ADI / FADI is further used to identify truly high-quality bird calls with diverse frequency structures.
[0051] Figure 4 The temporal distribution of the original segments, the top 10% of segments in the DAI ranking, and the final selected segments are shown. The results indicate that the original data exhibits a clear noise-dominated characteristic at different times of the day, while the top 10% of segments in the DAI ranking begin to show temporal clustering of bird calls. In particular, the final selected segments reveal a typical diurnal choral structure of birds: a significantly dense concentration of high-quality bird calls appears around dawn and sunset, while the selected results for daytime and nighttime periods are relatively sparse. This characteristic is highly consistent with the daily activity rhythms of birds in the ecological environment, and also indirectly verifies the effectiveness of the proposed selection method in real ecological soundscapes.
[0052] This invention proposes a rapid screening method for high-quality bird calls based on acoustic index fusion decision-making, applicable to bird call monitoring and ecoacoustic analysis under complex soundscape conditions in the wild. The method performs unified preprocessing and time segmentation on the collected environmental audio, and extracts multiple acoustic index features, including DCI, ADI, and FADI, through frequency domain analysis. Audio segments are then progressively screened using a hierarchical decision-making approach. First, DCI is used for rapid initial screening of the overall acoustic activity of audio segments to eliminate a large number of noise-dominated segments. Then, the frequency band distribution characteristics of candidate segments are further screened by combining the multi-index joint judgment of ADI and FADI, thereby improving the accuracy and robustness of screening high-quality bird call segments in complex noise backgrounds. The computational process of this invention is simple and efficient, suitable for large-scale offline data processing tasks. This method can accurately screen effective audio segments with high SNR and active bird calls from natural monitoring data under long-term, high background noise conditions, significantly improving the reliability and processing efficiency of subsequent acoustic analysis and index calculation. Meanwhile, this method has good adaptability, generalization ability and stability to different ecological regions and diverse natural soundscapes, and can provide a high-quality data foundation for bird species diversity assessment and the construction of a long-term ecological monitoring system, with good engineering application prospects and ecological research value.
Claims
1. A rapid screening method for high-quality bird calls based on acoustic index fusion decision, characterized in that, The steps are as follows: Step 1: Load and preprocess the acquired audio data to obtain audio data segments for analysis; Step 2: For each data segment in Step 1, estimate the average background noise power of the data segment, calculate the noise suppression complexity index which has a positive correlation with bird call activity, and complete the preliminary screening of candidate audio data segments accordingly. Step 3: For the candidate audio data segments obtained in Step 2, calculate two types of acoustic diversity indices to describe the frequency band energy distribution characteristics, and further filter the candidate audio data segments based on the frequency band activity characteristics. Step 4: Based on the acoustic diversity index and its corresponding frequency band features, construct adaptive discrimination conditions, perform joint judgment on audio data segments, and output high-quality bird sound segments.
2. The method for rapid screening of high-quality bird calls based on acoustic index fusion decision according to claim 1, characterized in that, The specific process of step 1 is as follows: Step 1-1: Within a preset date range, automatically load continuously acquired WAV audio files; based on the standard file naming rules adopted by the acquisition device, parse and extract the acquisition timestamp information corresponding to each audio file; Step 1-2: Sort the loaded audio files according to their chronological order based on the timestamp information obtained from the parsing. Steps 1-3: Perform uniform resampling on the sorted audio files and extract single-channel audio signals from the multi-channel audio; Steps 1-4: Divide the processed audio signal into multiple consecutive audio segments according to natural minutes, with each audio segment having a fixed duration of 1 minute; at the same time, record the actual acquisition time corresponding to each audio segment.
3. The method for rapid screening of high-quality bird calls based on acoustic index fusion decision according to claim 1, characterized in that, The specific process of step 2 is as follows: Step 2-1: Perform a short-time discrete Fourier transform on each audio data segment, and estimate the narrowband noise power at each frequency point based on the obtained time-frequency power spectrum; Step 2-2: Based on the narrowband noise power estimation results, calculate the noise suppression technical complexity index (DACI); Steps 2-3: Sort the DCI values corresponding to all audio segments to obtain a sequence of audio segments arranged from high to low acoustic complexity; Steps 2-4: Select audio segments whose DCI values are within a preset range from the sorting results as a set of candidate segments.
4. The method for rapid screening of high-quality bird calls based on acoustic index fusion decision according to claim 1, characterized in that, The specific process of step 3 is as follows: Step 3-1: For the candidate segment set obtained in Step 2, calculate the Acoustic Diversity Index (ADI) value for each audio segment; Step 3-2: For the candidate segment set obtained in Step 2, calculate the Acoustic Diversity Index (FADI) value based on frequency-varying threshold detection for each audio segment; Step 3-3: For the frequency band energy sequences corresponding to ADI and FADI respectively, sort the energy values of each frequency band from high to low; Steps 3-4: Calculate the cumulative energy of the sorted frequency band energy sequence; when the cumulative energy reaches a preset proportion of the total energy, determine the required number of frequency bands, denoted as strong_bands_ADI and strong_bands_FADI respectively; Steps 3-5: Determine whether strong_bands_ADI and strong_bands_FADI meet the preset frequency band quantity conditions. If not, discard the segment.
5. The method for rapid screening of high-quality bird calls based on acoustic index fusion decision according to claim 1, characterized in that, The specific process of step 4 is as follows: Step 4-1: For the filtered segments obtained in Step 3, using strong_bands_ADI and strong_bands_FADI as indicators of frequency band activity, introduce empirical adjustment coefficients. Dynamic discrimination thresholds T are constructed for ADI and FADI respectively. ADI T FADI : ； ； Step 4-2: Determine whether the ADI value and FADI value of the audio segment exceed the corresponding dynamic threshold at the same time. When the judgment conditions are met at the same time, the corresponding audio segment is judged as a high-quality bird sound segment and output as the final filtering result.
6. A high-quality bird call rapid screening system based on acoustic index fusion decision, characterized in that, The system for implementing the method according to any one of claims 1 to 5 comprises: The first module is used to load and preprocess the acquired audio data to obtain audio data segments for analysis. The second module estimates the average background noise power of each data segment obtained from the first module; calculates the noise suppression complexity index, which has a positive correlation with bird call activity, and completes the preliminary screening of candidate audio data segments accordingly. The third module calculates two types of acoustic diversity indices to describe the frequency band energy distribution characteristics for the candidate audio data segments obtained in the second module, and further filters the candidate audio data segments based on the frequency band activity characteristics. The fourth module constructs adaptive discrimination conditions based on the acoustic diversity index and its corresponding frequency band characteristics, performs joint judgment on audio data segments, and outputs high-quality bird sound segments.
7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of any of the methods described in claims 1-5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any one of claims 1-5.
9. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-5.