Back to Search Start Over

Birdcall Identification Using Mel-spectrum Based on ResNeSt50 Model

Authors :
Yihan Hong
Shizhen Liu
Chenhao Cui
Siyuan Lin
Source :
2021 IEEE International Conference on Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI).
Publication Year :
2021
Publisher :
IEEE, 2021.

Abstract

Audio processing and recognition technology is mainly to realize the recognition and prediction of audio data by converting audio data into image data by means of spectrum and other forms, and then performing network model training on the image data. This paper uses the ResNeSt50 model to classify and predict bird call based on a number of collected bird calls. We first preprocessed the raw data given by the material, and transformed the audio data into processed data that can be used by the model with the help of the Mel spectrogram. Then we use the ResNet50 network model to train and test the data set, and finally get the prediction result. Through experiments, we finally found that the ResNeSt50 model has better prediction performance than ResNet34, VGG19 and Simple NN models.

Details

Database :
OpenAIRE
Journal :
2021 IEEE International Conference on Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI)
Accession number :
edsair.doi...........f8bc2477e6a1216d9513fa4558a24f06
Full Text :
https://doi.org/10.1109/cei52496.2021.9574552