Saturday, October 5 2:00 pm – 4:30 pm
SESSION D: SIGNAL PROCESSING, PART 2
Chair: Robert Bristow-Johnson, Wave Mechanics, Burlington, VT, USA
D-1 Automatic Design of Sound Synthesis Techniques by Means of Genetic Programming—Ricardo A. Garcia, Chaoticom, Hampton Falls, NH, USA
Design of sound synthesis techniques (SST) is a difficult problem. It is usually assumed that it requires human ingenuity to find a suitable solution. Many of the SSTs commonly used are the fruit of experimentation and long refinement processes. An automated approach for design of SSTs is proposed. The problem is stated as a search in the multidimensional SST space. It uses genetic programming (GP) to suggest valid functional forms and standard optimization techniques to fit their internal parameters. A psychoacoustic auditory model is used to compute the perceptual distance between the target and test sounds. The developed AGeSS (automatic generator of sound synthesizers) system is introduced, and a simple example of the evolved SSTs is shown.
Convention Paper 5654
D-2 A Consumer Adjustable Dynamic Range Control System—Keith A. McMillen, Octiv, Inc., Berkeley, CA, USA
Advances in technology have afforded listeners an available dynamic range in excess of 120 dB. While impressive in proper concert halls and listening rooms, large dynamic ranges are not always realistic for all environments and musical styles. This paper describes a practical multi-band dynamics processor software object that can reside in low cost consumer products and allow the user to adjust dynamic range to fit his or her taste and listening environment.
Convention Paper 5655
D-3 A Simple, Efficient Algorithm for Reduction of Hiss Amplification Under High Dynamics Compression—Guillermo García, Creative Advanced Technology Center, Scotts Valley, CA, USA
We present a very simple, effective, and computationally efficient algorithm to reduce the typical hiss amplification (or breathing) artifact of dynamics compressors working under high compression ratios. The algorithm works in the time domain, is very easy to implement, has a very low computational cost, and requires little program memory, therefore being of special interest for consumer-audio applications.
Convention Paper 5656
D-4 Adaptive Predistortion Filter for Linearization of Digital PWM Power Amplifier Using Neural Networks—Minki Yang1, Jong-Hoon Oh2 - 1Pulsus Technologies, Inc., Seoul, Korea; 2Pohang University of Science and Technology, Pohang, Korea
The paper presents a method to compensate for nonlinear distortion of digital pulse width modulation (PWM) power amplifiers by pre-filtering the input signals using artificial neural networks. We first construct a model of the digital amplifier using artificial neural networks. Using this model, the artificial neural network model of a pre-distortion filter is trained such that the combined system, the digital amplifier, and pre-distortion filter produces an output that is linearly proportional to the input. The simulation results show that the artificial neural networks of the pre-distortion filter can effectively correct the nonlinear distortion of the digital amplifier.
Convention Paper 5657
D-5 Objective Measures of the Quality of Speech Transmission in a Real Mobile Network—Measuring, Estimate, and Prediction Method—Dragana Sagovnovic, Telekom Srbija, Belgrade, Yugoslavia
Contemporary means of communication (e.g., mobile telephony) have brought new limitations that telecom operators have to take into consideration. One of them is the fact that the types of deterioration of speech quality, perceived in mobile telephony, are different from the degradations noted in fixed telephony. This paper discusses, during the process of estimating the quality of speech transmission, the comparative (objective) methods for a formed database of degradations of a real mobile communications system. The consideration of the results of the objective-method tests is based on the development of a new objective method of speech quality. The results gathered during the comparison tests have been displayed and interpreted for different types of Serbian vowels: front vowels, /e/; mid vowels, /a/; and back vowels, /u/.
Convention paper 5658
