The Research On Audio Segementation Algorithm For Piano Tones

With the rapid development of Internet technology, the proportion of audio, video and other multimedia information in network is heavier and heavier, so the demands for multimedia information retrieval are. Audio information retrieval is an important branch of multimedia information retrieval, which is based primarily on the value of auditory feature.The premise of audio retrieval is audio recognition, one important step of which is structuring audio information.The structuralization of the audio information is to divid the audio into fragments of varying lengths based on criterion, with the signals in the same fragment have the same properties under this standard. Content information of the audio signal can be divided into three layers - the physical sample level, the acoustic feature level and the semantic level.The structuralization of the audio information is usually based on acoustic features.Through studying a large number of algorithms in content-based audio retrieval, the purpose of the existing segmentation algorithms is to split the audio clips into different types,but for the audio clip in the same type, which had a negative impact on the creation of more efficient audio identification, retrieval and score analysis.This article proposes a new audio segmentation algorithms based on weighted variance sum for piano tones. Combining the idea of audio segmentation algorithm based on BIC theory and information entropy, this algorithm could be used in the segmentation for the type of piano tones and the resut of which can even provide needed structural information for the semantic recognition and analysis.The algorithm has a remarkable effect on the pure piano tone played by single hand.For the piano music recorded in the natural environment, the algorithm can effectively segment each individual note, providing data base for music recognition and semantic analysis.However, in reality, practicing the piano is generally synchronously playing by hands.The preproceedings is necessary for the piano tone played synchronously by hands,which is called the blind source separation.In this article, the source separation algorithm ICA is introduced, and the different influence of commonly used non-liner function in the audio signal separation are analyzed.Experiments show that mix two piano tones played by single hand, then the result separated by ICA algorithm is basically the same with the recording audio before mixed.Thus it can be the preprocessing of the audio segmentation algorithm based on weighted variance sum, and the purpose of segmenting the piano tones by hands is achieved.