Real-time implementation of MELP vocoder

• Main Author: Teo, Tse Tsong.
• Other Authors: Tan, Eng Chong
• Format: Theses and Dissertations
• Language: English
• Published: 2008
• Subjects: DRNTU::Engineering::Computer science and engineering::Hardware::Input/output and data communications; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: http://hdl.handle.net/10356/13594
• Institution: Nanyang Technological University

The MELP (Mixed Excitation Linear Prediction) Vocoder has recently been accepted as the new Federal Standard (USA) after LPC-10 and CELP. This vocoder uses a mixture of noise and pulses in the synthesizer to simulate the residual signal obtained after LPC spectral whitening of the input speech. This pulse and noise mixture removes the characteristic buzziness from traditional LPC synthesized speech. In addition, a third voicing state is introduced that removes thumps and tonal noises. Spectral enhancement filtering and pulse dispersion filtering further enhance the synthesized voice quality. In this work, the TMS320C44 (running at 40MHz) is chosen as the DSP hardware platform for real-time implementation. The MELP algorithms are coded in C44 instruction set, making full use of its fast memories and registers. Efficient implementation of computationally intensive modules also contributes towards better performance and memory usage.