Separating musical audio signals

摘要

Musical audio source separation is an interesting and challenging area of research. Where we have the same number of microphones as sources to be separated, we can use methods based on independent component analysis (ICA). In many cases there are more sound sources than microphones, and for those cases we can use methods based on time-frequency masking, including the well-known non-negative matrix factorization (NMF) approach. However, this is still far from being a solved problem, particularly if we are looking for results with high perceived audio quality, suitable for upmixing or remixing applications, and currently may still require significant manual work to get the best results. Nevertheless, there are some promising areas of current research that suggest that progress is still possible, and it will be interesting to see how far we can go towards the target of fully automatic high quality audio source separation.