Research On Audio Scene Recognition Based On Sparse Decomposition

Now people have entered the information age, and with the development ofscience and technology, information is becoming increasingly important in people’susually living and production. Information displays in different ways in our lives, howto use it effectively is a significant mission that the information era entrusted to us.Within this, the audio signal is an indispensable part of one.Network multimedia and digital signal processing technology has now madegreat strides. Audio signal as part of the digital signal, its size is also growing.However, more and more audio data and that contain vast amounts of information,how to find the content we are interested in is not only a difficult problem, but also anurgent problem. Conventional manual search is clearly unable to effectively deal withthe multi-modal network, the massive scale of the audio data, so we need effectivetechnical means to achieve the audio data automatic processing and content analysisto identify. These technical means will support audio processing needs of varioustypes of intelligent systems in reality.Audio scene recognition is to detect and analysis of acoustic events in audiosignal, then understand the semantic content of the audio, and then to identify aparticular audio scene. The so-called audio scene is one of the audio data fragmentsthat contain the specific semantics, semantics often has a strong representation anddiscrimination is an important basis for human analysis and the difference betweenthe audio content and means, in reality there are many important applications.Audio scene recognition relies on the characteristics of the extracted audio signaland audio scene recognition mode to identify the characterization of the scene of thesemantic tags. The extraction of the audio signal characteristics with good effect is agreat aid to audio scene recognition. In the characteristics of the data analysis aspect,according to the ideological component analysis, the acoustic characteristics of theaudio signal can be optimized. Then, based on the idea of the component analysis, weuse a weighted method to optimize the characteristics. This paper uses the theory ofsparse decomposition to extract the sparse characteristics of audio signal. This kind offeature has long-time nature with good results in the audio scene recognition. We use the theory of component analysis to weight the characteristics to optimization. Byusing this method can acquire the key component of the audio signal characteristics.The sparse decomposition theory we use in this paper, is a new kind of signalprocessing originally applied to the image signal processing. When people use amethod of non-redundant orthogonal transformation in data representation, they findmany problems. Some of the signal itself is a complex mixture of signal. It cannot bewell expressed in a single orthogonal transformation. Sparse representation is to usethe atomic library that is a complete redundancy function system instead the basisfunctions for data representation. Atom is an element in the atomic library. Selectingm atoms in the atomic library, to express the original data signal, is data signal sparsedecomposition. Based on the signal sparse decomposition theory, a new model ofaudio scene recognition is put forward. First, training atom libraries over the targetscene and set outside scene and combination the two atom libraries. Then thecharacteristics of the audio signal will be recognized in the combination of atomiclibrary for sparse decomposition. Analysis whether the atoms get by sparsedecomposition belong to target audio scene atom library, and vote to determine whichscene this audio signal belongs to.The audio data using in the experiments in this paper, is right from the internet.The experimental results will also compare with the results get by the existing audioscene recognition model. To show more about the experiment and make theexperimental results convince. Then, experimental results were analyzed, andsummarized what we get. What this paper can continue to improve and perfect isprospect.