Automatic Recognition Of Animal Species Based On Voice

Automatic recognition of animal based on voice is very helpful in wildlife protection and other fields, this technology can effectively improve the recognition efficiency, productivity, saving manpower and resources. So, this paper proposed a system of automatic recognition of animal based on voice.Only Gunaseykaran and Revath published a paper yet, although this paper obtained great innovation and breakthrough, but his achievement can not applied to produce. The lower classification accuracy in the system leads to the low usability in reality, and then there are many shortcomings needed to be improved.As a representative of the type of traditional voice recognition technology, there are some drawbacks, such as it does not remove redundant data and invalid data in the link of original voice processing, there exists subjectivity in the link of feature selection, and the system does not remove the feature which is not suitable for zhe characteristics, all of those will reduce the computational efficiency and accuracy of the system.The traditional voice recognition process summarized as: pretreatment, feature extraction, training of the model. In view of the above shortcomings, this paper introduces a new voice recognition process: pretreatment, valid sound signal extraction, feature extraction, feature selection and training of the model. In the new processes, including a feature selection algorithm, outlier removal, the endpoint detection algorithm proposed in this paper, the aim of the three algorithms is to remove the invalid data and redundant data.In these new processes, including a feature selection algorithm, remove outliers and, a new algorithm. The main purpose of these algorithms has two, one is the purification of data as far as possible, remove the invalid data and redundant data; the second is to select features to make sure the features which are selected can represent those data better.The system introduced in this paper can achieve 90.84% recognition accuracy, and compared with the system of Gunasekar and Revathy work, the average accuracy rate has increased, which guarautee that the system is efficient enough for practicaldemands.Endpoint detection algorithm and outliers removal are part of the innovation of this article, therefore, this article also made four tests with the same data. The first test uses the two algorithms, the second test only use the the endpoint detection algorithm, the third test only use the outliers algorithm, the last test removed the two algorithms on the basis of the first test. Through the four tests, we can see the first test get the highest accuracy, and thelast test get the lowest accuracy. The four tests proves the effectiveness of these two algorithms.