An Audio Classification Approach Based on Machine Learning

The extraction of structured information and content semantics from audio is the key to deep processing of audio information, content-based retrieval and video analysis. This paper discusses the methods and techniques of audio event detection and semantic analysis in complex audio environment. Considering the number of non-target events, as well as problems such as detection strategy, detection time, the decision tree model based on AdaBoost is used to alleviate the sample imbalance problem. Then an integrated learning is used to provide a general framework for key audio event detection. The simulation results show that the multi-layer retrieval strategy effectively reduces false alarm and improves detection accuracy because of the low computational complexity of decision tree which also guarantees the detection time. At the same time, it shows better classification accuracy and speed for traditional audio recognition methods.