Análise comparativa em relação à qualidade e desempenho de codecs usando compressão por CELP e wavelets.

Abstract

This dissertation presents a comparative analysis between the opensource codec Speex (using the CELP algorithm) and two developed audio codecs based on the wavelet's theory. The implemented wavelets are respectively the Haar and the Coiflet wavelet with 30 coefficients. The application/scenario was a VoIP embedded device, thus both codecs have been compiled and ported for compatibility with the ARM920T's architecture. They have been evaluated according to the following criteria: signal's compression rate, quality at the output of the codec and the processing effort required for coding and decoding. To get the compression rate, the amount of bits at the codec's input and at the output were analyzed. The quality indexes were calculated using the MOS and the PESQ methods (subjective and objective respectively). By applying the QEmu simulator it has been possible to compare the codecs' performance (by means of tracing the data execution). According to the results of both evaluation methods, the signal quality of the Speex achieved better results. Speex is followed by Speex-Wavelet Coiflet and then Speex-Wavelet Haar. The compression rate has been higher for the Speex, that operated at 15kbps, comparing to 40kbps for the wavelet codecs. Regarding the processing overhead, the Speex-Wavelet Haar codec has executed less instructions and allocated less cycles of the processor, followed by the Speex-Wavelet Coiflet and then by the Speex.