PERFORMANCE OF LSF VECTOR QUANTIZERS FOR VSELP CODERS IN NOISY CHANNELS

Efficient quantization of synthesis filter coefficients for CELP (Code Excited Linear Prediction) coders is essential to achieve high qualit y speech at low rates. Three vector quantizers with good potential for utilization in low rate coders are studied. Each of them is implement ed inside the structure of the VSELP (Vector Sum Excited Linear Predic tion) coder, an important member in the class of CELP coders. These th ree vector quantizers are used to encode LSF (Line Spectral Frequency) parameters and are compared in terms of robustness to channel errors, complexity and quality of synthesized speech. Performance of synthesi zed speech is evaluated considering the objective measure of frequency weighted signal to noise ratio and subjective results obtained from l istening tests. With the purpose of improving the robustness to channe l errors, the application of simulated annealing to assign binary indi ces to the output levels of the quantizer is also investigated. A spli t vector quantization scheme which employs interframe prediction has s hown to be an attractive approach to encode the synthesis filter param eters. It provides a performance comparable to the IS-54 scheme and us es 10 bits less for each LSF frame.