Improving headphone spatialization for stereo music

Abstract

Music is mixed and mastered for playback on near- and far-field speakers, which presents a problem to the growing population of listeners who listen to music primarily on headphones. Playing legacy stereo mixes on headphones places the stereo image inside the listener's head and makes the image appear ultra-wide. While this is helpful for separating the center and stereo content, it has a detrimental effect on the spatialization of the music. It makes headphone listening unnatural and uncomfortable, although listeners have learned to accept it. In this work, we develop a system that processes stereo signals to provide a better sound image to headphone listeners. The sound image is improved by adding the necessary cues to the signal so as to externalize the perceived soundstage by making it similar to the soundstage experienced inside a mixing studio. In order to work with all headphones, our system tries to maintain the mastering equalization curve of the original stereo content. We employ head-related transfer functions, de-correlation, models of reverberation, and 1/3-octave-band equalization to realize our system. The results of pilot subjective evaluations suggest that our system makes music more natural and comfortable to listen to, although with some loss of quality.