Using digital signal processing to transcribe polyphonic music

The aim of the project described in this paper is to produce a system which could automatically convert recordings of musical performances to musical notation. The system should be able to handle multiple instruments playing simultaneously. Musicians often need to analyse and transcribe recorded performances, to reproduce or rearrange works for their own performances, or to find out how particular musical effects are achieved. This is a task requiring a good musical ear and plenty of practice. Very short or quiet notes, inner parts in large chordal voicings, parts written with the intention of being an inconspicuous harmonic or rhythmic background, etc. can all cause difficulties for the human ear. Using DSP for this process is a very obvious idea. All the required information is clearly present in the signal, frequency analysis techniques are well established and cheap processing power is available in the home PC market. Suitably priced such a product would probably sell well among musicians. There are a variety of commercial products on the market to notate or harmonise single melodic lines, but packages which could handle an entire band or orchestra do not seem to be available.