Impact of early reflection on real-time estimation of direction of arrival in different building environments

• Published in: Building and environment Vol. 147; pp. 434 - 443
• Main Authors: Seo, Jong Gak, Jeon, Jin Yong
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.01.2019; Elsevier BV
• Subjects: Algorithms; Background noise; Direction of arrival; Furniture; GCC-PHAT; Halls; Impact analysis; Indoor environments; Intelligibility; Localization; Microphone array; Noise levels; Real time; Reverberation time; Signal to noise ratio; Sound; Sound sources; Speech localization; Tracking; White noise; GCC-PHAT; Direction of arrival; Microphone array; Real-time; Reverberation time; Speech localization
• ISSN: 0360-1323; 1873-684X
• DOI: 10.1016/j.buildenv.2018.10.029

This study developed an 8-channel microphone array to track the location of a sound source in real time. In addition, a system that implements the GCC–PHAT function as the main algorithm was developed to measure the direction of arrival (DoA) in four spaces with different space volumes and reverberation times (RTs). The developed system exhibited usability in indoor environments with high background noises as it extracted DoA with errors lower than 3° when the signal-to-noise ratio (SNR) was higher than 5 dB for white noise. In addition, as is generally known, the accuracy of a localization system decreases as the space volume and RT increase. However, a higher DoA accuracy was observed in large spaces such as concert halls, in which the RT is high but early reflective surfaces are located far away, compared with that in spaces such as lecture rooms, in which the RT is relatively low and early reflections from nearby furniture and walls. Speech intelligibility was the highest when strong early reflections within 20 dB (compared with direct sound) were formed in the error range of ±3.2°. Furthermore, it was verified that the system was most effective when the frame length, which is a real-time sound source analysis parameter, was 5.8 ms to extract a small amount of operation and the amount of DoA information within the same time period through the evaluation results in the four spaces. •Investigated impacts of building environment on a DoA estimation.•Developed 8-channel linear microphone array and GCC-PHAT-based system.•Evaluated DOA characteristics according to space volume and reverberation.•Optimal frame length for the real-time system was selected as 5.3 ms•Strong early reflections within 20 dB have large impacts on speech intelligibility.