3-D Audio Synthesis: A DIY Approach for HRIR Database Acquisition

3-D audio synthesis is a computational method for generating and converting monaural sound sources into binaural sound, e.g., virtual spatial audio for headphones playback. Binaural synthesis requires the use of head-related impulse responses (HRIR), which can be measured or modeled, and are responsible for providing listeners with the sensation of sound spatialization. It is known that HRIRs are dependent on the anthropometric characteristics of individuals, thus, for a more realistic experience, it is best if the HRIRs are fitted to the listener’s own physiology. This paper aims to study a technique for capturing HRIRs using binaural in-ear microphones, and a computational approach for filtering out unwanted measurement artififacts, thus creating a personal HRIR database. By generating excitation signals and reproducing it through an audio monitor in an office-like environment, a raw database was first obtained, containing pairs of impulse responses (IR) for a few different points in space, varying the monitor’s horizontal and vertical angles with respect to the subject’s head, but with fixed head-position and distance radius. Since these measurements capture not only the subject’s HRIRs, but also the influence of the loudspeaker, the room and the microphones response, a linear predictive coding (LPC) filter was implemented, modeling approximate parametric IRs for these undesired components, thus enabling to filter them out. As a result, a small HRIR database was acquired, and used as a convolutional filter to virtualize monaural sound sources into 3-D audio. This measured database was then objectively compared with its filtered version, showing a superior signal-to-noise ratio (SNR). Subjectively, the acquired database was also compared with a public domain one, and a commercial virtual studio technology (VST) plugin, showing a fair level of satisfaction. Finally, a discussion about complexity of implementation versus satisfaction of results is presented, and further possibility of headphone equalization and head-tracking sensor integration as future work is proposed.