Rhythm Quantization for Transcription

摘要: Automatic Music Transcription is the extraction of an acceptable notation from performed music. One important task in this problem rhythm quantization which refers to categorization note durations. Although a pure mechanical performance rather straightforward, becomes increasingly difficult presence musical expression, i.e. systematic variations timing notes and tempo changes. For natural performances, we employ framework based on Bayesian statistics. We demonstrate that some simple schemata can be derived by assumptions about deviations. A general method, framework, vector (VQ). The algorithm operates short groups onsets thus flexible capturing structure deviations between neighboring performs better than rounding methods. Finally, present results examples.