Audio Coding Based on Long Temporal Segments: Experiments With Quantization of Excitation Signal

In this paper, we describe additional experiments based on a novel audio coding technique that uses an autoregressive model to approximate an audio signal's Hilbert envelope. This technique is performed over long segments (1000 ms) in critical-band-sized sub-bands. We have performed a series of experiments to find more efficient methods of quantizing the frequency components of the Hilbert carrier, which is the excitation found in the temporal audio signal. When using linear quantization, it was found that allocating 5 bits for transmitting the Hilbert carrier every 200 ms was sufficient. Other techniques, such as quantizing the first derivative of phase and using an iterative adaptive threshold, were examined.