Analysis of audio coding algorithms for networked embedded systems

Abstract

Today there are many different audio coding algorithms, some of them standardized by the ITU, like G.711, G.726 and G.728, and several other codecs by other organizations like Fraunhofer (mp3) and Microsoft (WMA). All these algorithms differ in several important features, like speech quality, bit or compression rate, robustness, delay, sampling frequency, complexity and range of use. The past decade we have witnessed a great progress towards application of low-rate speech/music coders as well as computer related voice/music applications. Central to this progress has been the development of speech coders capable of producing high-quality speech and music at low rates. Most of these coders incorporate mechanisms to: represent the spectral properties of speech/music and optimize the coder's performance for the human ear. The objective of this thesis is to compare audio coding algorithms for use in an embedded networked system: PCM and ADPCM for speech, and Mp3, WMA and Ogg-Vorbis for music. The thesis report start with a theoretical background of the different transport protocols, some signal theory and information about all the audio coding algorithms we would like to analyze. Numerical analysis with MathLab and a subjective assessment of audio quality are performed to find out which audio codec is best suited in an embedded VoIP system. Our main observations are; that the quantization noise increases when quantization levels decreases, that is, lower bit rate result in higher quantization noise; codecs implemented in MathLab with lower than 40 kbit/s bit rate is not suited for an audio based embedded system; and that WMA 48 kbit/s is very well suited for a networked embedded radio system.