Document Type
Thesis
Date of Award
5-31-1990
Degree Name
Master of Science in Electrical Engineering - (M.S.)
Department
Electrical and Computer Engineering
First Advisor
Ali N. Akansu
Second Advisor
Nirwan Ansari
Third Advisor
Zoran Siveski
Abstract
The Modified Hermite Transform (MHT), a suboptimum block transform scheme, is implemented on a TMS320C25 signal processing system with the objective of reducing the redundancy in speech signals and transmit a quantized vector representation at bitrates as less as 0.5 bits per sample.
The generation of the Hermite polynomials with a binomial and several first order IIR filter structures is presented focusing on the fact that no muliplications are required which simplifies the MHT algorithm considerably.
An assembler program performs forward and inverse transformation together with a vector quantization using adaptive bitallocation based on short-time statistics for the transform coefficients. Through several intermediate buffers the block oriented transform algorithm is converted into a pipeline structure, which is supported through the paging concept of the signal processor.
The performance of the MHT and DCT in terms of gain of coding over PCM and signal-to-noise ratio is evaluated in simulation using an AR(1) model and speech as source signals. The implementation is used to conduct intelligibility tests of transmitted voices with an average bitrate of 0.125 to 2 bits per sample.
Recommended Citation
Battenschlag, Peter, "An implementation of MHT algorithm for adaptive transform coding of speech" (1990). Theses. 2459.
https://digitalcommons.njit.edu/theses/2459
