Output details

Multiple-instrument polyphonic music transcription using a temporally constrained shift-invariant model.

Many recent approaches to automatic transcription are based on spectrogram factorisation, and model the observed data as the product of fixed spectral templates and a note activity matrix. This fails to account for the time-varying nature of musical tones, which in turn limits the quality of results. Our system is the first to model the temporal evolution (e.g. attack, decay, sustain) of tones, whilst also allowing for multiple simultaneous sources, multiple templates per source and pitch, and frequency variation of the notes. As a result, our transcription system outperforms all published results on three standard publicly available datasets.