Analysis and improvement of distortions due to listener position in stereo sound systems

The primary aim of stereo technology is to provide listeners with a two-dimensional auditory field in which the spatial characteristics of sound sources can be perceived. In stereo sound systems that utilise loudspeakers, accurate localisation of sound sources depends on the listener being positioned at what is commonly referred to as the sweet spot, which is regarded as the optimal listening position. The sweet spot is a relatively narrow region in which the listener receives the most precise spatial cues, typically defined by an equilateral triangle configuration between the listener and the two loudspeakers, both in terms of distance and angle. When the listener moves away from this optimal position, distortions occur in essential auditory attributes such as frequency, amplitude, and phase. Assuming the effects of room acoustics are negligible, these distortions arise from sound waves reaching the ears at different times, with varying amplitudes and frequency content. Such deviations in listener position lead to perceptual inaccuracies in the stereo image, making the localisation of sound sources more difficult and diminishing the overall quality of the stereophonic experience. Consequently, analysing and mitigating these spatial distortions has become a crucial topic of research in stereo audio reproduction. This dissertation aims to analyse listener position-induced distortions in stereo sound systems and propose corrective solutions to improve spatial accuracy. For this purpose, a custom-designed binaural microphone was constructed, calibrated, and employed in a full anechoic chamber to capture impulse responses across 169 distinct positions and head azimuth angles, including the sweet spot. These positions were systematically defined using 5 cm horizontal intervals and 15 degree azimuth steps. A logarithmic sine sweep signal was used for all measurements. Using a custom-developed software tool, the impulse responses obtained from off-axis locations were deconvolved with the reference data from the sweet spot, generating correction coefficients for each spatial position and head angle. These coefficients were then applied to the respective impulse responses, yielding corrected responses. The corrected responses were subsequently convolved with the original sine sweep signal to produce improved frequency responses, which were re-recorded at the same positions and angles. This procedure enabled a comparative evaluation of the corrected and uncorrected signals relative to the sweet spot data. The results demonstrate that position dependent distortions in both time and frequency domains can be mathematically modelled and significantly reduced through impulse response based correction algorithms. It was observed that in low-frequency bands, where interaural time difference (ITD) is the dominant localisation cue, higher correction accuracy was achieved by analysing the signals from each ear separately. Conversely, in high-frequency bands where interaural level difference (IID) prevails, analysing the average of both ears yielded better results. The proposed methods achieved error reduction rates of up to 80% in relevant frequency bands. Additionally, a custom digital signal processing (DSP) software was developed as part of this study, operating in parallel with color sensors that calculate distance and head orientation. This system dynamically adjusts the audio signals emitted from the loudspeakers in accordance with the listener's position and head azimuth. This software offers a functional solution for real-time correction based on listener position and orientation, enhancing spatial accuracy in stereo systems.