Automatic hexaphonic guitar transcription using non-negative constraints

摘要

Automatic music transcription is a widely studied problem, Typically, recordings that are used for transcription are taken from standard instruments, in the case of electric stringed instruments—such as the electric guitar—the recordings are captured from a standard pick-up, which unwantedly mixes the signals from each string and complicates subsequent analysis. We propose an approach to electric guitar transcription where the signal generated by each string at the guitar pickup is captured and analysed separately; thus providing six separate signals as opposed to one mixed signal, which enables finger positions to be identified. Such an instrument is known as a hexaphonic guitar and is a popular instrument for spatial music performances. We build the equipment necessary to modify a standard electric guitar into a hexaphonic guitar, and present an application of Non-Negative Matrix Factorisation to the task of transcription—where a basis for each note on the fretboard is learned and fitted to a magnitude spectrogram of the hexaphonic recording, which then undergoes a nonlinearity generating a piano roll representation of the music performance.