Collen, Patrice (2002) Bandwidth extension tools for audio digital signals. PhD thesis Traitement du signal, ENST.

Full text available as: - Bandwidth_Extension_Tools_Patrice_Collen_Nov_2002.pdf ( 1727 Kb )

Licence: Copyright

Abstract

To maintain a reasonable perceived quality and to reduce degradations, classical audio or speech source coding algorithms need to limit the audio bandwidth and to operate at low sampling rates. For a data rate of 20 kbit/s for instance, the bandwidth of audio signals doesn't exceed 6kHz with classical audio coders. The signals produced suffer from some quality degradation due to the lack of high energy components.
To overcome this problem, new methods for improving the quality of bandlimited signal are proposed in this document. With the use of a little more transmitted information, bandwidth extension tools allow the recovery of spectral highband components and thus enhance the quality of such bandlimited signals.
The method exploits signal redundancy in the spectral domain and uses lowband components to synthesise the fullband signal.
This thesis was completed through a contract with France Telecom R&D. The project aim was to design an effective and low bitrate bandwidth extension tool.
The PAT (Perceptual Audio Transposition) technology produced during these 3 years was proposed in DRM (Digital Radio Mondiale) and MPEG-4 (Moving Picture Experts Group) consortiums.
This document is split into four parts.
Based on psychoacoustics properties and characteristics of treated audio signals, the first part introduces bandwidth extension tools. According to these considerations, a first bandwidth extension scheme is introduced. The process of bandwidth extension can be divided in two independent tasks: the extension of the frequency components (high frequency regeneration) and the highband spectral envelope estimation.
The second part is dedicated to estimation, coding and adjustment of the spectral envelope. Two particular techniques are considered: The first method is based on linear prediction, and the second method consists of modelling the spectral envelope in the frequency domain.
The third part examines in detail several techniques for high frequency regeneration. In particular, spectral translations in the frequency domain and non-linear distortions are developed.
Finally, in the fourth part, a new bandwidth extension scheme is proposed. Subjective tests evaluate the performances of this technique in the context of MPEG-4 normalisation.
A new method for highband audio compression techniques is introduced in this document. For same quality, subjective tests demonstrate that the bitrate reduction is about 25%. This new bandwidth extension tool demonstrates high performance in audio coding.

Statistiques de consultation

Repository Staff Only: edit this item