high performance computing on graphics processing units: hgpu.org

Three dimensional finite difference time domain schemes can be used as an approach to spatial audio simulation. By embedding a model of the human head in a 3D computational space, such simulations can emulate binaural sound localisation. This approach normally relies on using high sample rates to give finely detailed models, and is computationally intensive. This paper examines the use of head models within audio rate FDTD schemes, ranging from 176.4 down to 44.1 kHz. Using GPU computing with Nvidia’s CUDA architecture, simulations can be accelerated many times over a serial computation in C. This allows efficient, dynamic simulations to be produced where sounds can be moved around during the runtime. Sound examples have been generated by placing a personalised head model inside an anechoic cube. At the lowest sample rate, 44.1 kHz, localisation is clear in the horizontal plane but much less so in the other dimensions. At 176.4, there is far greater three dimensional depth, with perceptible front to back, and some vertical movement.