Sound Localization for Humanoid Robots - Building Audio-Motor Maps based on the HRTF

A survey on sound source localization in robotics: From binaural to array processing methods

“…Such an idea was successfully assessed in [29] through a dedicated neural network able to generalize learning to new acoustic conditions. One can also cite [30], or [31], where the iCub humanoid robot's head was endowed with two pinnae. The localization is performed by mapping the aforementioned sound features to the corresponding location of the source through a learning method.…”

“…Reproducing such capabilities in Robotics is a difficult problem due to the lack of a model of the pinnae shapes which lead to elevation dependent notches. Yet, as a rule of thumb, these shapes must be irregular or asymmetric, and artificial pinnae were proposed in [37], [38], [31], [35], or [39]. Fig.…”

Multimodal Language Acquisition Based on Motor Learning and Interaction

“…The auditory sensor consist of two microphones and artificial pinnas [26]. To get a more compact representation of the sound signal it is transformed into a tonotopic sound representation (sound features).…”

Variational EM for binaural sound-source separation and localization

“…We note that the vast majority of current SSL approaches mainly focus on a rough estimation of the azimuth, or one-dimensional (1D) localization [9,5,10,7], and that very few perform 2D localization [8]. Alternatively, some approaches [6,11,12] bypass the explicit mapping model and perform 2D localization using an exhaustive search in an HRTF look-up table. However, this process is unstable and hardly scalable in practice as the number of required associations yields too prohibitive memory and computational costs.…”