Measurement of Relative Transfer Function for Own Voice in Head-Mounted Microphone Array

In this paper, we investigate a suitable vocalization method for measuring the relative transfer function (RTF) of a user’s own voice in head-mounted microphone arrays (HMMAs). The realization of auditory devices that can extract a desired voice from a mixture is of great interest. Blind source separation can separate the target speaker’s voice, but it also removes the user’s own voice, making conversation difficult. Our aim is to estimate the RTF of own voice in advance to prevent such removal and improve dialogue quality. For RTF measurements using a loudspeaker, a broadband signal such as a time-stretched pulse (TSP) signal is typically used as the source signal. However, when measuring the RTF of own voice, it is not possible to produce the TSP signal. To address this, we investigate which vocalizations are suitable for RTF measurement. We evaluate the consistency of estimated RTF from three types of vocalization using a self-fabricated HMMA. Our results show that vocalization with smooth pitch variation is more suitable for RTF measurement than others. Furthermore, the results of individual dependency on RTFs obtained from male and female subjects showed little individual dependence in the low-frequency band.