Approaches and databases for online calibration of binaural sound localization for robotic heads

Abstract

In this paper, we evaluate adaptive sound localization algorithms for robotic heads. To this end we built a 3 degree-of-freedom head with two microphones encased in artificial pinnae (outer ears). The geometry of the head and pinnae induce temporal differences in the sound recorded at each microphone. These differences change with the frequency of the sound, location of the sound, and orientation of the robot in a complex manner. To learn the relationship between these auditory differences and the location of a sound source, we applied machine learning methods to a database of different audio source locations and robot head orientations. Our approach achieves a mean error of 2.5 degrees for azimuth and 11 degrees for elevation for estimating the position of an audio source. The impressive results highlight the benefits of a two-stage regression model to make use of the properties of the artificial pinnae for elevation estimation. In this work, the algorithms were trained using ground truth data provided by a motion capture system. We are currently generalizing the approach so that the training signal is provided online based on a real-time face detection and speech detection system.

Bi-Direction Interaural Matching Filter and Decision Weighting Fusion for Sound Source Localization in Noisy Environments

“…This superiority is because HIMF makes a weighting fusion by the cost function based on FIMF and BIMF in the decision level, which improves the performance of localization and reduces the systematic error. Figure 4 compares the azimuth localization with the state-of-the-art methods, including TDC [8], hierarchical system [9], online calibration [10] and probabilistic model [2] for different SNRs. Specifically, all the methods perform well, reaching more than 90%, in quiet circumstance.…”

Binaural Sound Localization Based on Reverberation Weighting and Generalized Parametric Mapping

“…3) Azimuth estimation in noisy environments: Some comparisons with several state-of-the-art methods, including a classical Hierarchical System [18], Online Calibration [19] and Interaural Matching Filter (IMF) [44], are carried out in the noisy environments without reverberation, i.e., T R = 0 s. In fact, [18] and [19] belong to the hierarchical methods using different binaural cues. In [44], the IMF was proposed as a new localization cue, which contains somewhat relative transfer function information, to achieve a real-time sound localization in noisy environments.…”

“…The experimental office room is of dimensions (6 m × 5 m × 3 m). Since the walls and roof of the room are made of painted concrete and the floor is resilient, the reverberation time of this room is [30] 81.43% 83.58% 87.26% Raspaud et al [24] 76.28% 77.57% 81.93% IMF [44] 80.13% 82.26% 85.65% Online Calibration [19] 71.29% 74.52% 75.34% Hierarchical System [18] 77.21% 79.67% 83.49% • at an interval of 10 • . For each direction, 20 groups of audio data are recorded, which are from the speech data of 10 males and 10 females in the TIMIT database [45].…”

“…For instance, Li et al proposed a three-layer binaural SSL system based on the Bayesian rule [18]. Along with the similar hierarchical architecture, 2 experiments in [19] demonstrated that the hierarchical system can effectively reduce the time consumption of SSL. For acoustic interference, most existing methods focus on the extraction of robust binaural cues, while ignoring their valid combination.…”

Binaural sound source localization based on generalized parametric model and two-layer matching strategy in complex environments

“…For instance, Li et al proposed a three-layer hierarchical binaural sound source localization system based on Bayes-Rule [15]. Along with the similar hierarchical architectures like Finger et al [16], experiments showed that hierarchical system could reduce time consumption effectively. Willert et al introduced a probabilistic model for binaural sound source localization by extracting binaural cues from cochleagrams generated by a cochlear model [17].…”