Signal requantization is applied whenever the word length of audio signals needs to be reduced. This is for instance the case when an audio signal needs to be stored on CD, and was originally available at a higher precision (like the pro-standard 24 bit). Unfortunately, signal requantization inevitably leads to a reduction of the audio precision. In analogy with perceptual audio coding (like MP3), it is possible to make the signal requantization minimally audible by appropriate shaping of the spectrum of the error signal. The specifications for this optimal shaping filter can be obtained from a model for auditory perception. However, these specifications are time varying as they depend on the masking properties of the audio signal. Consequently, adaptive psycho-acoustic noise shaping could not been applied in real-time in the past because of the high computational complexity of both the psycho-acoustic modelling, and the filter design proper. In our own research, we were able to show that the filter could be obtained by solving a simple linear Least Squares problem, thereby substantially reducing the computational complexity. In this project, we intent to study whether we can also reduce the computational complexity of the psychoacoustic module. We think we will be able to achieve this by adapting the perceptual weighting functions, that are typically used in GSM, from speech at 8 kHz sampling frequency to audio at 44.1 kHz.
Effective start/end date1/01/0431/12/05

    Flemish discipline codes

  • Electrical and electronic engineering
  • Mathematical sciences
  • Basic sciences

    Research areas

  • Digital Audio Processing, Audio Coding, Adaptive Signal Processing, Digital Signal Processing, Auditory Perception

ID: 3015486