AES London 2010
Paper Session P1
P1 - High Performance Audio Processing
Saturday, May 22, 09:30 — 12:30 (Room C3)
Chair: Neil Harris, New Transducers Ltd. (NXT) - Cambridge, UK
P1-1 Model-Driven Development of Audio Processing Applications for Multi-Core Processors—Tiziano Leidi, ICIMSI-SUPSI - Manno, Switzerland; Thierry Heeb, Digimath - Sainte-Croix, Switzerland; Marco Colla, ICIMSI-SUPSI - Manno, Switzerland; Jean-Philippe Thiran, EPFL - Lausanne, Switzerland
Chip-level multiprocessors are still very young and available forecasts anticipate a strong evolution for the forthcoming decade. To exploit them, efficient and robust applications have to be built with the appropriate algorithms and software architectures. Model-driven development is able to lower some barriers toward applications that process audio in parallel on multi-cores. It allows using abstractions to simplify and mask complex aspects of the development process and helps avoid inefficiencies and subtle bugs. This paper presents some evolutions of Audio n-Genie, an open-source environment for model-driven development of audio processing applications, which has been recently enhanced with support for parallel processing on multi-cores.
Convention Paper 7961 (Purchase now)
P1-2 Real-Time Additive Synthesis with One Million Sinusoids Using a GPU—Lauri Savioja, NVIDIA Research - Helsinki, Finland, Aalto University School of Science and Technology, Espoo, Finland; Vesa Välimäki, Aalto University School of Science and Technology - Espoo, Finland; Julius O. Smith III, Stanford University - Palo Alto, CA, USA
Additive synthesis is one of the fundamental sound synthesis techniques. It is based on the principle that each sound can be represented as a superposition of sine waves of different frequencies. That task can be done fully parallel and thus it is suitable for GPU (graphics processing unit) implementation. In this paper we show that it is possible to compute over one million unique sine waves in real-time using a current GPU. That performance depends on the applied buffer sizes, but close to the maximum result is reachable already with a buffer of 500 samples.
Convention Paper 7962 (Purchase now)
P1-3 A GPGPU Approach to Improved Acoustic Finite Difference Time Domain Calculations—Jamie A. S. Angus, Andrew Caunce, University of Salford - Salford, Greater Manchester, UK
This paper shows how to improve the efficiency and accuracy of Finite Difference Time Domain acoustic simulation by both calculating the differences using spectral methods and performing these calculations on a Graphics Processing Unit (GPU) rather than a CPU. These changes to the calculation method result in an increase in accuracy as well as a reduction in computational expense. The recent advances in the way that GPU’s are programmed (for example using CUDA on Nvidia's GPU) now make them an ideal platform on which to perform scientific computations at very high speeds and very low power consumption.
Convention Paper 7963 (Purchase now)
P1-4 Digital Equalization Filter: New Solution to the Frequency Response Near Nyquist and Evaluation by Listening Tests—Thorsten Schmidt, Cube-Tec International - Bremen, Germany; Joerg Bitzer, Jade-University of Applied Sciences - Oldenburg, Germany
Current design methods for digital equalization filter face the problem of a frequency response increasingly deviating from their analog equivalent close to the Nyquist frequency. This paper deals with a new way to design equalization filters, which improve this behavior over the entire frequency range between 0 Hz (DC) and Nyquist. The theoretical approach is shown and examples of low pass, peak-, and shelving-filters are compared to state-of-the-art techniques. Listening tests were made to verify the audible differences and rate the quality of the different design methods.
Convention Paper 7964 (Purchase now)
P1-5 Audio Equalization with Fixed-Pole Parallel Filters: An Efficient Alternative to Complex Smoothing—Balázs Bank, Budapest University of Technology and Economics - Budapest, Hungary
Recently, the fixed-pole design of parallel second-order filters has been proposed to accomplish arbitrary frequency resolution similarly to Kautz filters, at 2/3 of their computational cost. This paper relates the parallel filter to the complex smoothing of transfer functions. Complex smoothing is a well-established method for limiting the frequency resolution of audio transfer functions for analysis, modeling, and equalization purposes. It is shown that the parallel filter response is similar to the one obtained by complex smoothing the target response using a hanning window: a 1/ß octave resolution is achieved by using ß/2 pole pairs per octave in the parallel filter. Accordingly, the parallel filter can be either used as an efficient implementation of smoothed responses, or, it can be designed from the unsmoothed responses directly, eliminating the need of frequency-domain processing. In addition, the theoretical equivalence of parallel filters and Kautz filters is developed, and the formulas for converting between the parameters of the two structures are given. Examples of loudspeaker-room equalization are provided.
Convention Paper 7965 (Purchase now)
P1-6 Rapid and Automated Development of Audio Digital Signal Processing Algorithms for Mobile Devices—David Trainor, APTX - Belfast, N. Ireland, UK
Software applications and programming languages are available to assist audio DSP algorithm developers and mobile device designers, including Matlab/Simulink, C/C++, and assembly languages. These tools provide some assistance for algorithmic experimentation and subsequent refinement to highly-efficient embedded software. However, a typical design flow is still highly iterative, with manual software recoding, translation, and optimization. This paper introduces a software libraries and design techniques that integrate existing commercial audio algorithm design tools and permit intuitive algorithmic experimentation and automated translation of audio algorithms to efficient embedded software. These techniques have been incorporated into a new software framework, and the operation of this framework is described using the example of a custom audio coding algorithm targeted to a mobile audio device.
Convention Paper 7966 (Purchase now)