Design And Implementation Of Acoustic Scene Detection Mechanism

The main task of acoustic scene detection mechanism is to annotate an audio stream which contains a variety of acoustic scene automatically,and it combines signal processing and pattern recognition technology. It plays an important role in many applications, such as audio and video retrieval. Acoustic scene detection can also be used as the front-end application of speech recognition and voiceprint recognition. The main tasks of this paper is as follows:(1) Implemented and optimized two basic acoustic scene detection systems, then, the performance test of the basic system is completed;(2) Optimized the system further more. Parameter adaptation are implemented. According to the matched model and threshold of current test data, the system's performance is improved. New spectral features are used and combined with current basic system, and improve the performance of the system for noise distortion; Musical signals are further classified by its music key; Lastly, designed and implemented a prototype system of acoustic scene detection system based on NMF feature, and verify that the performance of system is improved by adding NMF features to an MFCC-based system;(3)Implemented the noise reduction algorithm based on non-negative matrix factorization(NMF). After comparing the algorithm of spectral subtraction and NMF, they are integrated into the current basic system, and the performance of the system is improved;(4) Designed and implemented the online acoustic scene detection system based on the basic system. Combining with the characteristics of online system, Implemented online voice activity detection. The performance of the system is improved by updating the threshold and model;...