09:00am-10:30am
<SP2> Signal Processing, PART2

SP2-1
A perceptual measure for audio source separation
Mingu Lee, Insuk Heo, Keong-Mo Sung, Seoul National University, Seoul, Korea, Nakjin Choi, Samsung Electronics, Suwon, Korea

In this paper, an improved method on evaluating the performance of blind audio source separation is discussed. Based on Vincent s measures, defined by SDR, SIR, SNR and SAR [1], several well-known psychoacoustic characteristics, such as equal-loudness contours and masking effect, are considered to make these measures more relevant to human auditory system. Correlation between the results of the proposed methods and those of carefully designed listening tests is presented for verification.

SP2-2
Tiny DSP: DSP Core, Algorithm Development and 'Device Mastering
Nathan Bentall, Peter Eastty, Duncan Stott, Oxford Digital, Oxfordshire, UK

Market expectations of small size, low cost and, in many cases, a requirement for very low power create difficult challenges in the electric and acoustic design of consumer devices that include loud speakers such as mobile phones, laptop computers and flat panel displays. Developments in the field of 'sound improvement' algorithms can go a long way to improving the listening experience, often making an attempt at acoustic correction, but even on high ticket-price items, fierce competition results in very high sensitivity to component cost, which can rule out many DSP devices; rapid development is mandatory due to
development cost and large variety of new models. A combination of processor design and tool set are described, which simultaneously addresses these issues; an implementation of a commercially available device is described; an example usage is outlined for improvement of a consumer-device; a real time parameter adjustment tool is presented, which enables real-time 'tuning' of developed algorithms to facilitate the ultimate aim: a combination of low cost DSP and algorithm which can be rapidly individually tailored to any given device - a process described here as 'Device Mastering'.

SP2-3
Simple High-band Extension Method Using Wavelet for Mobile Device
Sang-keun Oh, LG Electronics, Seoul, Korea

This paper suggests the simple higher-frequency band extension (HBE) method based on digital wavelet transform (DWT). The method estimates the band missed through the perceptual coding process by using DWT filters. As the input signal for processing, we use the band passed signal wave (f) including the missed band which is used as the approximation coefficients for estimating. Then the method synthesizes the band passed signal wave (f ) using DWT and reconstructs the output signal. In order to estimate the missed band from the approximation coefficients, we design the FIR type inverse filter of DWT analysis filter in integer numbers. So we expect the method can behave in real time for variable type of mobile devices such as mobile handsets and mp3 players using ARM9 core as their processing units.

11:00am-12:30pm
<3DS> 3-D Audio and Synthetic Audio

3DS-1
Robust Crosstalk Cancellation Based on Eenergy-density Control
Young-Cheol Park, Junho Lee, Dae-Hee Youn, Yonsei University, Seoul, Korea

This paper presents a robust crosstalk cancellation algorithm based on the control of acoustic energy density. Since it is known that energy density distribution in a space is more uniform than acoustic pressure, the control of energy density can provide more stable crosstalk cancellation than the case of controlling squared pressure. Thus, the proposed algorithm produces a robust crosstalk cancellation in the vicinity of error sensor. Simulation results confirm that the use of proposed algorithm provide stable zone of crosstalk cancellation, which contributes to the robustness in relation to movement of listener.

3DS-2
Designing Low-Dimensional Interaction for Mobile Navigation in 3D Audio Spaces+C13
Till Schafers, Michael Rohs, Sascha Spors, Alexander Raake, Jens Ahrens, Deutsche Telekom Laboratories, TU Berlin, Germany

In this paper we explore spatial audio as a new design space for applications like teleconferencing and audio stream management on mobile devices. Especially in conjunction with input techniques using motion-tracking, the interaction has to be thoroughly designed in order to allow low-dimensional input devices like gyroscopic sensors to be used for controlling the rather complex spatial setting of the virtual audio space. We propose a new interaction scheme that allows the mapping of low-dimensional input data to navigation of a listener within the spatial setting.

3DS-3
A Consonance-Maximization Tuning Algorithm in Equal-Temperament Synthesized Tones
Sang Bae Chon, Sang Ha Park, Koeng-Mo Sung, Seoul National University, Seoul, Korea, Ah Jin Yim, Ewha Woman's University, Seoul, Korea

We propose a tuning system for sound synthesis using equal temperament that maximizes the consonance of the chords. By removing dissonance (such as in beating and roughness) and rearranging the harmonic structure, it is possible to achieve more consonant sound, which is only possible in computer based sound synthesis. To verify the performance of the proposed algorithm, listening tests were carried out with the most popular chords, i.e. perfect 5th, major 3rd and major 6th, on synthesized piano, violin and flute sounds. They confirmed that the modified major 3rd and major 6th chords using the proposed algorithm sounded more harmonic than the original ones. We believe that the harmonicity and consonance can be improved in commercial applications such as ring tone synthesis.

01:30pm-03:00pm
<Workshop> Audio in IT Industry