Mixed structural models for 3D audio in virtual environments

In the world of Information and communications technology (ICT strategies for innovation and development are increasingly focusing on applications that require spatial representation and real-time interaction with and within 3D-media environments. One of the major challenges that such applications have to address is user-centricity, reflecting e.g. on developing complexity-hiding services so that people can personalize their own delivery of services. In these terms, multimodal interfaces represent a key factor for enabling an inclusive use of new technologies by everyone. In order to achieve this, multimodal realistic models that describe our environment are needed, and in particular models that accurately describe the acoustics of the environment and communication through the auditory modality are required. Examples of currently active research directions and application areas include 3DTV and future internet, 3D visual-sound scene coding, transmission and reconstruction and teleconferencing systems, to name but a few. The concurrent presence of multimodal senses and activities make multimodal virtual environments potentially flexible and adaptive, allowing users to switch between modalities as needed during the continuously changing conditions of use situation. Augmentation through additional modalities and sensory substitution techniques are compelling ingredients for presenting information non-visually, when the visual bandwidth is overloaded, when data are visually occluded, or when the visual channel is not available to the user (e.g., for visually impaired people). Multimodal systems for the representation of spatial information largely benefit from the development of audio engines that implement extensive knowledge of spatial hearing and virtual acoustics. Models for spatial audio can provide accurate dynamic information about the relation between the sound source and the surrounding environment, including the listener and his/her body which acts as an additional filter. Indeed, this information cannot be substituted by any other modality (i.e., visual or tactile). Nevertheless, today’s spatial representation of audio within sonification tends to be simplistic and with poor interaction capabilities, being multimedia systems currently focused on graphics processing mostly, and integrated with simple stereo or multi-channel surround-sound. On a much different level lie binaural rendering approaches based on headphone reproduction, taking into account that possible disadvantages (e.g. invasiveness, non-flat frequency responses) are counterbalanced by a number of desirable features. Indeed, these systems might control and/or eliminate reverberation and other acoustic effects of the real listening space, reduce background noise, and provide adaptable and portable audio displays, all of which are relevant aspects especially in innovative contexts. Most of the binaural sound rendering techniques currently exploited in research rely on the use of Head-Related Transfer Functions (HRTFs), i.e. peculiar filters that capture the acoustic effects of the human head, torso and ears. HRTFs allow loyal simulation of the audio signal that arrives at the entrance of the ear canal as a function of the sound source’s spatial position. HRTF filters are usually presented under the form of acoustic signals acquired on dummy heads built according to mean anthropometric measurements or actual listeners. Nevertheless, anthropometric features of the human body have a key role in HRTF shaping: several studies have attested how listening to non-individual binaural sounds results in evident localization errors. On the other hand, individual HRTF measurements on a significant number of subjects are both time- and resource-expensive. Several techniques for the design of synthetic HRTFs have been proposed during the last two decades and the most promising one relies on structural HRTF models. In this revolutionary approach, the most important effects involved in spatial sound perception (acoustic delays and shadowing due to head diffraction, reflections on pinna contours and shoulders, resonances inside the ear cavities) are isolated and modeled separately with a corresponding filtering element. HRTF selection and modeling procedures can be determined by physical interpretation: parameters of each rendering blocks or selection criteria can be estimated from real and simulated data and related to anthropometric geometries. Effective personal auditory displays represent an innovative breakthrough for a plethora of applications and structural approach can also allow for effective scalability depending on the available computational resources or bandwidth. Scenes with multiple highly realistic audiovisual objects are easily managed exploiting parallelism of increasingly ubiquitous Graphics Processing Units (GPUs). Building individual headphone equalization with perceptually robust inverse filtering techniques represents a fundamental step towards the creation of personal virtual auditory displays (VADs). To this regard, several applications might benefit from these considerations: multi-channel downmix over headphones, personal cinema, spatial audio rendering in mobile devices, computer-game engines and individual binaural audio standards for movie and music production, to name but a few. This thesis presents a family of approaches that overcome the current limitations of headphone-based 3D audio systems, aiming at building personalized VADs through structural binaural audio models for an immersive sound reproduction. The resulting models allow for an interesting form of content adaptation and personalization, since they include parameters related to the user’s anthropometry in addition to those related to the sound sources and the environment. The covered research directions converge to a novel framework for the design and customization of synthetic HRTFs that combines the structural modeling paradigm with other HRTF selection techniques (inspired by non-individualized HRTF selection procedures) and represents the main novel contribution of this thesis: the Mixed Structural Modeling (MSM) approach considers the global HRTF as a combination of structural components, which can be chosen to include either synthetic or recorded components. In both cases, customization is based on individual anthropometric data, which are used to either fit the model parameters or to select a measured/simulated component within a set of available impulse responses. The definition and experimental validation of the MSM approach addresses several pivotal issues towards the acquisition and delivery of acoustic scenes and designing guidelines for personalized 3D audio virtual environments holding the potential of novel forms of customized communication and interaction with sound and music contents. The thesis also presents a multimodal interactive system which is used to conduct subjective test on multi-sensory integration in virtual environments. Four experimental scenarios are proposed in order to test the capabilities of auditory feedback jointly to tactile or visual modalities. 3D audio feedback related to user’s movements during simple target following tasks is tested as an applicative example of audio-visual rehabilitation system. Results from perceptual experiments of direction of footstep sounds interactively generated during walking and provided through headphones highlight how spatial information can clarify the semantic congruence between movement and multimodal feedback. A real-time, physically informed, audio-tactile interactive system encodes spatial information in the context of virtual map presentation with particular attention to orientation and mobility (O&M) learning processes addressed to visually impaired people. Finally, an experiment analyzes the haptic estimation of size of a virtual 3D object (a stair-step) whereas the exploration is accompanied by a real-time generated auditory feedback whose parameters vary as a function of the height of the interaction point. The collected data from these experiments suggest that well-designed multimodal feedback, exploiting 3D audio models, can definitely be used to improve performance in virtual reality and learning processes in orientation and complex motor tasks, thanks to the high level of attention, engagement, and presence provided to the user. The research framework, based on the MSM approach, serves as an important evaluation tool with the aim of progressively determining the relevant spatial attributes of sound for each application domain. In this perspective, such studies represent a novelty in the current literature on virtual and augmented reality, especially concerning the use of sonification techniques in several aspects of spatial cognition and internal multisensory representation of the body. This thesis is organized as follows. An overview of spatial hearing and binaural technology through headphones is given in Chapter 1. Chapter 2 is devoted to the Mixed Structural Modeling formalism and philosophy. In Chapter 3, topics in structural modeling for each body component are studied, previous research and two new models, i.e. near-field distance dependency and external-ear spectral cue, are presented. Chapter 4 deals with a complete case study of the mixed structural modeling approach and provides insights about the main innovative aspects of such modus operandi. Chapter 5 gives an overview of number of a number of proposed tools for the analysis and synthesis of HRTFs. System architectural guidelines and constraints are discussed in terms of real-time issues, mobility requirements and customized audio delivery. In Chapter 6, two case studies investigate the behavioral importance of spatial attribute of sound and how continuous interaction with virtual environments can benefit from using spatial audio algorithms. Chapter 7 describes a set of experiments aimed at assessing the contribution of binaural audio through headphones in learning processes of spatial cognitive maps and exploration of virtual objects. Finally, conclusions are drawn and new research horizons for further work are exposed in Chapter 8.