In an intelligent editing environment, the semantic music structure can be used as beneficial assistance during the post production process. In this paper we propose a new approach to extract both low and high level hierarchical structure from vocal tracks of multi-track master recordings. Contrary to most segmentation methods for polyphonic audio, we utilize extra information available when analyzing a single audio track. A sequence of symbols is derived using a hierarchical decomposition method involving onset detection, pitch tracking and timbre modelling to capture phonetic similarity. Results show that the applied model well captures similarity of short voice segments.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.