CN116524939A - An automatic identification method of bird song species based on ECAPA-TDNN - Google Patents

Abstract

The invention discloses an automatic identification method of birdsong species based on ECAPA‑TDNN, which includes: collecting birdsong segments, performing preprocessing and feature extraction, and obtaining Mel frequency cepstral coefficients; sending Mel frequency cepstral coefficients into ECAPA ‑TDNN network for training; through the speech endpoint detection algorithm based on Gaussian mixture model, the silent segment is removed, and the segment containing the bird's song is extracted; the Mel frequency cepstral coefficient corresponding to the bird's song segment is obtained, and the trained model is input for recognition. Get the results; display the recognition results one by one on the GUI, perform quantitative statistics by category, draw a spectrum diagram, and display the recognition information in the export table. The present invention improves the accuracy rate in the bird song classification scene through the ECAPA-TDNN model, realizes the automatic classification and preliminary analysis of long segment data of bird song, reduces the workload of manual tailoring, and facilitates subsequent in-depth analysis in the ecological field .

Description

An automatic identification method of bird song species based on ECAPA-TDNN

technical field

The invention belongs to the technical field of acoustic monitoring and audio signal recognition, in particular to an ECAPA-TDNN-based automatic bird song species recognition method.

Background technique

As an important part of the ecosystem, birds provide an important basis for ecologists to understand biodiversity and climate change in a region. Passive acoustic monitoring has the characteristics of low cost, wide range, and non-invasiveness, making bird song signals an important data source for monitoring bird activities. With the deterioration of the ecological environment and the difficulties in governance in recent years, species identification, behavior analysis, and acoustic index research based on bird songs have great application value.

At present, the species identification algorithms based on birdsong mainly include: 1) classification methods based on template matching, such as dynamic time planning template algorithm, which has the problems of large amount of calculation and low calculation efficiency; 2) traditional machine learning algorithm, Such as random forest, support vector machine, hidden Markov model, etc., this method is greatly affected by noise, and has high requirements for the signal-to-noise ratio of the data set; 3) Use deep learning for species identification, such as Alex Net, VGG16 model, etc. It is currently popular, but it is rarely applied to species identification based on bird songs in China.

The ECAPA-TDNN model was proposed in 2020. The introduction of the SE (squeeze excitation) module and the channel attention mechanism enables the model to learn more global information in the audio data, and has become the mainstream voiceprint model. The open source voiceprint recognition system released by Baidu's Paddle Speech uses ECAPA-TDNN to extract voiceprint features, and the recognition error rate is as low as 0.95%.

Therefore, a big problem in the prior art is that the field of bird song species recognition lacks a verified mainstream neural network model with high accuracy. In addition, the current bird song recognition algorithm needs to manually intercept clips containing only bird songs for prediction. When inputting long-term bird song audio, the manual workload is heavy and there is observer bias. In-depth analysis and follow-up research bring inconvenience.

Contents of the invention

The purpose of the present invention is to address the problems in the above-mentioned prior art, to provide a method for automatic species identification of birdsong based on ECAPA-TDNN, to realize automatic segmentation and species identification of singing audio while improving the accuracy of identification.

The technical solution that realizes the object of the present invention is: a kind of bird song species automatic recognition method based on ECAPA-TDNN, described method comprises the following steps:

Step 1. Collect bird song signals and preprocess them to construct a bird song data set, and then obtain Mel frequency cepstral coefficients through feature extraction;

Step 2, input ECAPA-TDNN network with described mel frequency cepstral coefficient and train, obtain ECAPA-TDNN birdsong classification model;

Step 3, input bird singing audio, remove the silent segment and extract the segment containing bird singing through the speech endpoint detection algorithm based on Gaussian mixture model;

Step 4, based on the segment containing birdsong, extract the Mel-frequency cepstral coefficients according to the method in step 1, input the ECAPA-TDNN birdsong classification model for recognition, and obtain the recognition result.

Further, the preprocessing described in step 1 specifically includes:

Step 1-1-1, for the birdsong signal, eliminate the DC component and perform pre-emphasis;

Step 1-1-2, perform high-pass filtering;

Step 1-1-3, carry out sub-frame processing;

Step 1-1-4, use the Hanning window for windowing.

Further, the pre-emphasis in step 1-1-1 is specifically realized by a first-order FIR high-pass digital filter whose transfer function is H(z)=1-az ^-1 , where a is the pre-emphasis coefficient.

Further, the Mel-frequency cepstral coefficients are obtained through feature extraction as described in step 1, which specifically includes:

Step 1-2-1, perform short-time Fourier transform on the bird song signal in the bird song data set, take the absolute value of the obtained value, and then square it to obtain the energy spectrum;

Step 1-2-2, constructing a Mel filter bank, and performing a dot product operation with the energy spectrum to obtain a Mel spectrogram;

Step 1-2-3, taking the logarithm of the mel spectrogram;

Step 1-2-4, performing discrete cosine transform, and taking the first P data to obtain Mel-frequency cepstral coefficients; said P is an integer.

Further, the ECAPA-TDNN network includes a convolutional neural layer, three SE-Res2Block layers, an Attentive Statistics Pooling layer and a fully connected layer;

The Mel-frequency cepstral coefficients are input into the ECAPA-TDNN network model for training, and the specific process includes:

Input the Mel-frequency cepstral coefficients into the convolutional neural layer to obtain potential audio features;

Perform multi-layer feature fusion of the potential audio features through the SE-Res2Block layer to extract global information;

Concatenate the outputs of the three SE-Res2Block layers according to the feature dimension;

Through the Attentive Statistics Pooling (statistical pooling with attention mechanism) layer, the mean and standard deviation based on the attention mechanism are obtained, and the vector is obtained by concatenation according to the feature dimension;

Carrying out softmax classification to the vector through a fully connected layer to obtain a classification result;

Based on the classification results, the cross-entropy loss function is used to update the network parameters through back propagation, and the ECAPA-TDNN bird song classification model is obtained.

Further, step 3 specifically includes the following process:

Step 3-1, preprocessing and framing the bird singing audio, wherein the preprocessing process is the same as step 1, creating multiple classes, and dividing the bird singing audio into segments and storing them in the class;

Step 3-2, mute judgment, specifically includes:

Step 3-2-1, divide the segment into subbands, and calculate the logarithmic energy of the subbands;

Step 3-2-2, when the total energy of the frame is greater than the minimum energy required to trigger the audio signal, for each subband, calculate the probability P(X|H1) of the speech based on the speech mixture Gaussian model, and calculate the probability P(X|H1) based on the noise mixture Gaussian The probability of the noise of the model P(X|H0);

Step 3-2-3, calculating the likelihood ratio of the subband based on the above two probabilities: likelihood ratio=log(P(X|H1)/P(X|H0));

Step 3-2-4, if the likelihood ratio of one of the subbands meets the preset threshold, it is judged as a voiced segment;

and or, accumulating the likelihood ratio of each subband as the overall likelihood ratio, and if the overall likelihood ratio meets the preset threshold, it is judged as a voiced segment;

Otherwise, it is judged as a silent segment;

Step 3-3, the collection of audio clips, specifically includes:

Step 3-3-1, create a data container with two ends, add an object, and get the number of voiced classes;

Step 3-3-2, when the number of audio clips is greater than 90% of the capacity of the data container, it is judged that the birdsong starts, and the data in the current data container is written into the constructed empty list;

Step 3-3-3, repeat step 3-3-1 and step 3-3-2, when the number of mute segments is greater than 90%, end the list writing;

Step 3-3-4, return the list data, that is, the segment containing only the sound of birdsong.

Further, step 3-2-1 specifically includes: according to the spectral characteristics and energy distribution of birdsong, each frame of birdsong signal is divided into 200-2000Hz, 2000-3000Hz, 3000-3500Hz, 3500-4500Hz, 4500-8000Hz, 8000-24000Hz six sub-bands, calculate sub-band energy and total energy, and take logarithm.

Further, the method also includes:

Step 5: Display the recognition results one by one on the GUI, conduct quantitative statistics by category, draw a spectrum diagram, and display the recognition information in the export table, including the start and end time of bird song, recognition results and similarity.

Further, the specific process of step 5 includes:

Step 5-1, build a graphical user interface, display the segment index, recognition results, and similarity one by one, and perform quantitative statistics by bird song type;

Step 5-2, draw the spectrogram of the input bird singing audio;

Step 5-3, create a new table to store the start and end time of birdsong, recognition results, and similarity information.

Compared with the prior art, the present invention has the remarkable advantages of:

1) Compared with traditional machine learning, deep learning can quickly and accurately learn audio potential features. The ECAPA-TDNN model emphasizes channel attention mechanism and multi-layer feature fusion. Experimental results show that the accuracy of bird song species classification is significantly improved.

2) The present invention is based on the mute detection algorithm designed by the Gaussian mixture model, which realizes the automatic cutting of the singing disc segment, with high accuracy and good segmentation effect.

3) The graphical user interface designed by the present invention is convenient and intuitive, and the user can independently select the audio to be identified, and the functions of statistics, drawing and tables realize the preliminary analysis and display of the audio to be identified.

The present invention will be described in further detail below in conjunction with the accompanying drawings.

Description of drawings

Fig. 1 is the flow chart of the automatic identification method of bird song species based on ECAPA-TDNN in the present invention.

Figure 2 is the structure diagram of ECAPA-TDNN.

Figure 3 is the structure diagram of SE-Res2Block of ECAPA-TDNN.

Fig. 4 is a flowchart of silence detection based on Gaussian mixture model.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

It should be noted that if there is a directional indication (such as up, down, left, right, front, back...) in the embodiment of the present invention, the directional indication is only used to explain the position in a certain posture (as shown in the accompanying drawing). If the specific posture changes, the directional indication will also change accordingly.

In addition, if there are descriptions involving "first", "second" and so on in the embodiments of the present invention, the descriptions of "first", "second" and so on are only for descriptive purposes, and should not be interpreted as indicating or implying Its relative importance or implicitly indicates the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include at least one of these features. In addition, the technical solutions of the various embodiments can be combined with each other, but it must be based on the realization of those skilled in the art. When the combination of technical solutions is contradictory or cannot be realized, it should be considered that the combination of technical solutions does not exist , nor within the scope of protection required by the present invention.

In one embodiment, in conjunction with Fig. 1, a kind of bird song species automatic identification method based on ECAPA-TDNN is provided, and described method comprises the following steps:

Step 1, collect bird song signals and perform preprocessing to form a relatively pure bird song data set, and then obtain Mel frequency cepstral coefficients through feature extraction;

Described pretreatment comprises:

Step 1-1-1, for the birdsong signal, eliminate the DC component and perform pre-emphasis, specifically through a first-order FIR high-pass digital filter with a transfer function of H(z)=1-az ^-1 , where a is pre-emphasis Coefficient, a=0.97, to enhance the high frequency component in the signal;

Step 1-1-2, perform high-pass filtering; since speech noise and environmental noise are mainly concentrated below 350Hz, the signal is passed through an 8-order Butterworth high-pass filter with a cutoff frequency of 350Hz to obtain a relatively pure bird sound signal;

Step 1-1-3, perform frame division processing, the frame length is 2048, and the frame shift is 512;

In step 1-1-4, use a Hanning window to add a window to eliminate the discontinuity between frames after framing.

The feature extraction to obtain the Mel frequency cepstral coefficients specifically includes:

Step 1-2-1, perform short-time Fourier transform on the bird song signal in the bird song data set, take the absolute value of the obtained value, and then square it to obtain the energy spectrum;

Step 1-2-2, constructing a Mel filter bank, the number of filters is 128, and performing a dot product operation with the energy spectrum to obtain a Mel spectrogram;

Step 1-2-3, taking the logarithm of the mel spectrogram;

In step 1-2-4, discrete cosine transform is performed, and the first 80 data are taken to obtain Mel-frequency cepstral coefficients; the P is an integer.

Step 2, the Mel-frequency cepstral coefficient input ECAPA-TDNN network (emphasis channel attention, delay neural network of propagation and aggregation) is trained, obtains ECAPA-TDNN birdsong classification model; In conjunction with Fig. 2, described ECAPA -TDNN network includes convolutional neural layer, three SE-Res2Block layers, Attentive Statistics Pooling (statistical pooling with attention mechanism) layer and fully connected layer. The specific process of this step includes:

Step 2-1, the Mel-frequency cepstral coefficient passes through the convolutional neural layer to obtain potential audio features;

In conjunction with Figure 3, step 2-2, the potential audio features are fused with multi-layer features through the SE-Res2Block layer to extract global information;

Step 2-3, connect the outputs of the three SE-Res2Block layers in series according to the feature dimension;

Steps 2-4, through the Attentive Statistics Pooling layer, get the mean and standard deviation based on the attention mechanism, concatenate according to the feature dimension, and get a 3072-dimensional vector;

Step 2-5, performing softmax classification on the vector through a fully connected layer to obtain a classification result;

Steps 2-6, based on the classification results, use the cross-entropy loss function to update the network parameters through backpropagation to obtain the ECAPA-TDNN bird song classification model.

Step 3: Input 1 minute of bird singing audio, and use the Gaussian mixture model-based speech endpoint detection algorithm to remove the silent segment and extract the segment containing bird singing. Combined with Figure 4, it specifically includes:

Step 3-1, preprocessing and framing the bird singing audio, wherein the preprocessing process is the same as step 1, creating multiple classes, dividing the bird singing audio into segments every 100ms and storing them in the class;

Step 3-2, mute judgment, specifically includes:

Step 3-2-1, divide the segment into subbands, and calculate the logarithmic energy of the subbands; divide each frame of bird song signal into 200-2000Hz, 2000-3000Hz, 3000-3500Hz according to the frequency spectrum characteristics and energy distribution of birdsong , 3500-4500Hz, 4500-8000Hz, 8000-24000Hz six sub-bands, calculate the sub-band energy and total energy, and take the logarithm;

Step 3-2-2, when the total energy of the frame is greater than the minimum energy required to trigger the audio signal, for each subband, calculate the probability P(X|H1) of the speech based on the speech mixture Gaussian model, and calculate the probability P(X|H1) based on the noise mixture Gaussian The probability of the noise of the model P(X|H0);

Step 3-2-3, calculating the likelihood ratio of the subband based on the above two probabilities: likelihood ratio=log(P(X|H1)/P(X|H0));

Step 3-2-4, if the likelihood ratio of one of the subbands meets the preset threshold, it is judged as a voiced segment (local);

Or, accumulate the likelihood ratio of each sub-band as the overall likelihood ratio, and if the overall likelihood ratio meets the preset threshold, it is judged as a voiced segment (global); here, as long as one of the global and local is satisfied, it is a voiced segment;

Otherwise, it is judged as a silent segment;

Step 3-3, the collection of audio clips, specifically includes:

Step 3-3-1, create a data container with two ends, add an object (instance of Frame class, mute detection result), and obtain the number of classes with sound;

Step 3-3-2, when the number of audio clips is greater than 90% of the capacity of the data container, it is judged that the birdsong starts, and the data in the current data container is written into the constructed empty list;

Step 3-3-3, repeat step 3-3-1 and step 3-3-2, when the number of mute segments is greater than 90%, end the list writing;

Step 3-3-4, return the list data, that is, the segment containing only the sound of birdsong.

Step 4, based on the segment containing birdsong, extract the Mel-frequency cepstral coefficients according to the method in step 1, input the ECAPA-TDNN birdsong classification model for recognition, and obtain the recognition result.

Step 5: Display the recognition results one by one on the GUI, perform quantitative statistics by category, draw a spectrum diagram, and display the recognition information in the export table, including the start and end time of bird song, recognition results and similarity, etc. Specifically include:

Step 5-1, build a graphical user interface, display the segment index, recognition results, and similarity one by one, and perform quantitative statistics by bird song type;

Step 5-2, draw the spectrogram of the input bird singing audio;

Step 5-3, create a new Excel table to store the start and end time of birdsong, recognition results, similarity information, etc.

In one embodiment, a kind of bird song species automatic recognition system based on ECAPA-TDNN is provided, and said system comprises:

The first module is used for preprocessing and feature extraction of birdsong signals to obtain Mel frequency cepstral coefficients;

The second module is used to train the ECAPA-TDNN network based on Mel frequency cepstral coefficients to obtain the ECAPA-TDNN bird song classification model;

The third module is used for silent detection, removing silent segments, and generating a data set containing only bird song segments;

The fourth module is used for preprocessing and feature extraction of segmented bird song segments, and recognizes them through the ECAPA-TDNN bird song classification model;

The fifth module is used to realize user interaction and display classification and analysis results.

For the specific limitations of the ECAPA-TDNN-based automatic bird song species identification system, please refer to the above-mentioned limitations on the ECAPA-TDNN-based bird song species automatic identification method, and will not be repeated here. Each module in the above-mentioned ECAPA-TDNN-based bird song species automatic identification system can be fully or partially realized by software, hardware and combinations thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, and can also be stored in the memory of the computer device in the form of software, so that the processor can invoke and execute the corresponding operations of the above-mentioned modules.

In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and operable on the processor. When the processor executes the computer program, the following steps are implemented:

Step 1. Collect bird song signals and preprocess them to construct a bird song data set, and then obtain Mel frequency cepstral coefficients through feature extraction;

Step 2, input ECAPA-TDNN network with described mel frequency cepstral coefficient and train, obtain ECAPA-TDNN birdsong classification model;

Step 3, input bird singing audio, remove the silent segment and extract the segment containing bird singing through the speech endpoint detection algorithm based on Gaussian mixture model;

Step 4, based on the segment containing birdsong, extract the Mel-frequency cepstral coefficients according to the method of step 1, input the ECAPA-TDNN birdsong classification model to identify, and obtain the recognition result;

Step 5: Display the recognition results one by one on the GUI, conduct quantitative statistics by category, draw a spectrum diagram, and display the recognition information in the export table, including the start and end time of bird song, recognition results and similarity.

For the specific limitations of each step, please refer to the above-mentioned limitations on the automatic identification method of bird song species based on ECAPA-TDNN, and will not be repeated here.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

Step 1. Collect bird song signals and preprocess them to construct a bird song data set, and then obtain Mel frequency cepstral coefficients through feature extraction;

Step 2, input ECAPA-TDNN network with described mel frequency cepstral coefficient and train, obtain ECAPA-TDNN birdsong classification model;

Step 3, input bird singing audio, remove the silent segment and extract the segment containing bird singing through the speech endpoint detection algorithm based on Gaussian mixture model;

Step 4, based on the segment containing birdsong, extract the Mel-frequency cepstral coefficients according to the method of step 1, input the ECAPA-TDNN birdsong classification model to identify, and obtain the recognition result;

Step 5: Display the recognition results one by one on the GUI, conduct quantitative statistics by category, draw a spectrum diagram, and display the recognition information in the export table, including the start and end time of bird song, recognition results and similarity.

For the specific limitations of each step, please refer to the above-mentioned limitations on the automatic identification method of bird song species based on ECAPA-TDNN, and will not be repeated here.

The invention is convenient and fast, has strong practicability and high accuracy, fully combines the advantages of deep learning, and realizes the automatic segmentation and species identification of bird songs, which is of great significance for the study of biodiversity in ecosystems and the protection of endangered birds.

The basic principles, main features and advantages of the present invention have been shown and described above. Those skilled in the art should understand that the present invention is not limited by the above-mentioned embodiments. The above-mentioned embodiments and descriptions only illustrate the principles of the present invention. Within the spirit and principles, any modifications, equivalent replacements, improvements, etc., shall be included in the protection scope of the present invention.

Claims (10)

1. a kind of bird song species automatic identification method based on ECAPA-TDNN, it is characterized in that, described method comprises the following steps: Step 1. Collect bird song signals and preprocess them to construct a bird song data set, and then obtain Mel frequency cepstral coefficients through feature extraction; Step 2, input ECAPA-TDNN network with described mel frequency cepstral coefficient and train, obtain ECAPA-TDNN birdsong classification model; Step 3, input bird singing audio, remove the silent segment and extract the segment containing bird singing through the speech endpoint detection algorithm based on Gaussian mixture model; Step 4, based on the segment containing birdsong, extract the Mel-frequency cepstral coefficients according to the method in step 1, input the ECAPA-TDNN birdsong classification model for recognition, and obtain the recognition result. 2. the bird song species automatic recognition method based on ECAPA-TDNN according to claim 1, is characterized in that, the preprocessing described in step 1 specifically comprises: Step 1-1-1, for the birdsong signal, eliminate the DC component and perform pre-emphasis; Step 1-1-2, perform high-pass filtering; Step 1-1-3, carry out sub-frame processing; Step 1-1-4, use the Hanning window for windowing. 3. the bird's song species automatic recognition method based on ECAPA-TDNN according to claim 2, is characterized in that, pre-emphasis is specifically passed through transfer function H (z)=1-az ^-1 in the step 1-1-1 First-order FIR high-pass digital filter implementation, where a is the pre-emphasis coefficient. 4. the bird song species automatic identification method based on ECAPA-TDNN according to claim 2, is characterized in that, described in the step 1 by feature extraction, obtains Mel frequency cepstral coefficient, specifically comprises: Step 1-2-1, perform short-time Fourier transform on the bird song signal in the bird song data set, take the absolute value of the obtained value, and then square it to obtain the energy spectrum; Step 1-2-2, constructing a Mel filter bank, and performing a dot product operation with the energy spectrum to obtain a Mel spectrogram; Step 1-2-3, taking the logarithm of the mel spectrogram; Step 1-2-4, performing discrete cosine transform, and taking the first P data to obtain Mel-frequency cepstral coefficients; said P is an integer. 5. the bird song species automatic identification method based on ECAPA-TDNN according to claim 1, is characterized in that, described ECAPA-TDNN network comprises convolution neural layer, three SE-Res2Block layers, Attentive Statistics Pooling layer and full connection layer; The Mel-frequency cepstral coefficients are input into the ECAPA-TDNN network model for training, and the specific process includes: Input the Mel-frequency cepstral coefficients into the convolutional neural layer to obtain potential audio features; Perform multi-layer feature fusion of the potential audio features through the SE-Res2Block layer to extract global information; Concatenate the outputs of the three SE-Res2Block layers according to the feature dimension; Through the Attentive Statistics Pooling layer, the mean and standard deviation based on the attention mechanism are obtained, and the vector is obtained by concatenation according to the feature dimension; Carrying out softmax classification to the vector through a fully connected layer to obtain a classification result; Based on the classification results, the cross-entropy loss function is used to update the network parameters through back propagation, and the ECAPA-TDNN bird song classification model is obtained. 6. the bird song species automatic identification method based on ECAPA-TDNN according to claim 1, is characterized in that, step 3 specifically comprises the following process: Step 3-1, preprocessing and framing the bird singing audio, wherein the preprocessing process is the same as step 1, creating multiple classes, and dividing the bird singing audio into segments and storing them in the class; Step 3-2, mute judgment, specifically includes: Step 3-2-1, divide the segment into subbands, and calculate the logarithmic energy of the subbands; Step 3-2-2, when the total energy of the frame is greater than the minimum energy required to trigger the audio signal, for each subband, calculate the probability P(X|H1) of the speech based on the speech mixture Gaussian model, and calculate the probability P(X|H1) based on the noise mixture Gaussian The probability of the noise of the model P(X|H0); Step 3-2-3, calculating the likelihood ratio of the subband based on the above two probabilities: likelihood ratio=log(P(X|H1)/P(X|H0)); Step 3-2-4, if the likelihood ratio of one of the subbands meets the preset threshold, it is judged as a voiced segment; Or, accumulating the likelihood ratio of each subband as the overall likelihood ratio, and if the overall likelihood ratio meets the preset threshold, it is judged as a voiced segment; Otherwise, it is judged as a silent segment; Step 3-3, the collection of audio clips, specifically includes: Step 3-3-1, create a data container with two ends, add an object, and get the number of voiced classes; Step 3-3-2, when the number of audio clips is greater than 90% of the capacity of the data container, it is judged that the birdsong starts, and the data in the current data container is written into the constructed empty list; Step 3-3-3, repeat step 3-3-1 and step 3-3-2, when the number of mute segments is greater than 90%, end the list writing; Step 3-3-4, return the list data, that is, the segment containing only the sound of birdsong. 7. the bird song species automatic identification method based on ECAPA-TDNN according to claim 6, is characterized in that, step 3-2-1 specifically comprises: according to bird song spectrum characteristic and energy distribution, every frame bird song signal Divide into six subbands of 200-2000Hz, 2000-3000Hz, 3000-3500Hz, 3500-4500Hz, 4500-8000Hz, 8000-24000Hz, calculate subband energy and total energy, and take logarithm. 8. the bird song species automatic identification method based on ECAPA-TDNN according to claim 6, is characterized in that, described method also comprises: Step 5, display the recognition results one by one on the GUI, perform quantitative statistics by category, draw a spectrum diagram, and display the recognition information in the export table, including the start and end time of bird song, recognition results and similarity. 9. the bird song species automatic identification method based on ECAPA-TDNN according to claim 8, is characterized in that, step 5 specific processes comprise: Step 5-1, build a graphical user interface, display the segment index, recognition results, and similarity one by one, and perform quantitative statistics by bird song type; Step 5-2, draw the spectrogram of the input bird singing audio; Step 5-3, create a new table to store the start and end time of birdsong, recognition results, and similarity information. 10. based on the bird song species automatic identification system based on ECAPA-TDNN of any one described method of claim 1 to 9, it is characterized in that, described system comprises: The first module is used for preprocessing and feature extraction of birdsong signals to obtain Mel frequency cepstral coefficients; The second module is used to train the ECAPA-TDNN network based on Mel frequency cepstral coefficients to obtain the ECAPA-TDNN bird song classification model; The third module is used for silent detection, removing silent segments, and generating a data set containing only bird song segments; The fourth module is used for preprocessing and feature extraction of segmented bird song segments, and recognizes them through the ECAPA-TDNN bird song classification model; The fifth module is used to realize user interaction and display classification and analysis results.