Blind source separation of musical signals applied to selectable-listening-point audio reconstruction

Kenta Niwa, Takanori Nishino, Chiyomi Miyajima, Kazuya TakedaThe Journal of the Acoustical Society of America 2006 · DOI

Abstract

An audio reproduction system is being developed to generate binaural signals at user-selected listening positions from audio signals of multiple sound sources recorded with multiple microphones. The system consists of a sound decomposer based on blind source separation (BSS) and a sound recomposer using head-related transfer functions (HRTFs). In this work, musical signals of multiple instruments are separated into the original independent signals of the respective instruments using BSS based on frequency-domain independent component analysis (FD-ICA). HRTFs are then convoluted to the separated sound sources to reconstruct the music sounds of the selected listening point. Preliminary experiments are conducted using two or three sound sources of instruments: flute, violin, and piano generated by a synthesizer. The synthesizer sounds are played through small directional loudspeakers or spherical omnidirectional loudspeakers in a low-reverberant sound-proof room or moderately reverberant lecture room and recorded with multiple microphones. Live music played on real instruments including flute, violin, and trumpet are also recorded in the same rooms. Audio signals of respective instruments obtained by BSS are available on our web site.
BibTeX
@article{niwa2006blind,
  author = {Kenta Niwa and Takanori Nishino and Chiyomi Miyajima and Kazuya Takeda},
  title = {{Blind source separation of musical signals applied to selectable-listening-point audio reconstruction}},
  journal = {The Journal of the Acoustical Society of America},
  year = {2006},
  doi = {10.1121/1.4788145},
}

Loading related content...

Figure