DSP Implementation of a Perceptual Speech Coder

Project Type: Master/Diploma Thesis
Student: Rauter Markus


 The goal of modern speech coders is to achieve a reconstruction quality as good as possible, while keeping the data rate and computational complexity as low as possible. This work presents a perceptual speech coder whose model is based on the human auditory system and which is therefore not exclusively restricted to the processing of speech signals. On the basis of a given Matlab simulation, the coder was implemented on a TSM320C6711 DSP from Texas Instruments. The main focus was thereby set on the analysis- and synthesis filter bank. Two different types have been realised and compared concerning their memory- and computing time requirements: a 20-channel FIR Gammatone (GT), and an oversampled Frequency-Warped Transform (FWT) Filter Bank. The comparision shows that the FWT filter bank is more efficient concerning both, memory- and computing time requirements.