Automated evaluation of acoustic talker direction finder algorithms in the varechoic chamber

Abstract

Acoustic talker direction finders have potential applications to camera pointing for teleconferencing and to microphone array beam steering for audio communication and voice processing systems. This paper describes a laboratory setup and computer interface that was developed for testing talker direction finder algorithms in the Bell Labs Varechoic chamber, a room with computer-controlled absorbing panels. The procedure exploits the full capability of the facility by automatically stepping through a sequence of room panel configurations, outputting a digital speech signal, running the processor, and collecting the data. The advantage of this technique is that it allows for testing under a multitude of different acoustic conditions in the same physical location, thereby enabling a general characterization of the algorithm under evaluation. As an example of the technique, we have implemented the Fischell–Coker talker direction finder algorithm using real-time C-code running on an SGI workstation, which is the same machine that is used to orchestrate the automatic testing procedure.

Adaptive eigenvalue decomposition algorithm for passive acoustic source localization

“…For FC, a 100 ms window was used. 10 For the new method, we have used the UFLMS 24 with ϭ0.003. The length of the adaptive vector u is taken as Lϭ2M ϭ512.…”

“…9 and was tested in a real time implementation in Ref. 10. The principle of this approach is to identify the onset of pitch periods and to cross-correlate these within a certain acceptance window.…”

Acoustic MIMO Signal Processing

“…The four panel configurations are 0%, 30%, 75%, and 89% open. During these measurements the north wall behind the microphone array was covered by a three-inch-thick fiberglass pillow, so that the reverberation time was smaller than reported in [234] for the same panel configuration. (Also, there has been some deterioration over the years due to loss of oil between panels.)…”

Audio-visual tracking for natural interactivity

“…We adopt a variant of the cross correlation technique that was developed by [4] and implemented by [17] to determine the talker location. The incoming signal at each of the two microphones is processed in several steps.…”