POLYPHONIC MUSIC INFORMATION RETRIEVAL BASED ON MULTI-LABEL CASCADE CLASSIFICATION SYSTEM
Abstract
Recognition and separation of sounds played by various instruments is very useful in labeling audio files with semantic information. This is a non-trivial task requiring sound analysis, but the results can aid automatic indexing and browsing music data when searching for melodies played by user specified instruments. Melody match based on pitch detection technology has drawn much attention and a lot of MIR systems have been developed to fulfill this task. However, musical instrument recognition remains an unsolved problem in the domain. Numerous approaches on acoustic feature extraction have already been proposed for timbre recognition. Unfortunately, none of those monophonic timbre estimation algorithms can be successfully applied to polyphonic sounds, which are the more usual cases in the real music world. This has stimulated the research on multi-labeled instrument classification and new features development for content-based automatic music information retrieval. The original audio signals are the large volume of unstructured sequential values, which are not suitable for traditional data mining algorithms; while the acoustical features are sometime not sufficient for instrument recognition in polyphonic sounds because they are higher-level representatives of raw signal lacking details of original information. In order to capture the patterns which evolve on the time scale, new temporal features are introduced to supply more temporal information for the timbre recognition. We will introduce the multi-labeled classification system to estimate multiple timbre information from the polyphonic sound by classification based on acoustic features and short-term power spectrum matching. In order to achieve higher estimation rate, we introduced the hierarchically structured cascade classification system under the inspiration of the human perceptual process. This cascade classification system makes a first estimate on the higher level decision attribute, which stands for the musical instrument family. Then, the further estimation is done within that specific family range. Experiments showed better performance of a hierarchical system than the traditional flat classification method which directly estimates the instrument without higher level of family information analysis.Traditional hierarchical structures were constructed in human semantics, which are meaningful from human perspective but not appropriate for the cascade system. We introduce the new hierarchical instrument schema according to the clustering results of the acoustic features. This new schema better describes the similarity among different instruments or among different playing techniques of the same instrument. The classification results show the higher accuracy of cascade system with the new schema compared to the traditional schemas. The query answering system is built based on the cascade classifier.