Artificial Bandwidth Extension Using H∞ Optimization and Speech Production Model
Deepika Gupta, Hanumant Singh Shekhawat · 2019
This work presents a new method for artificial bandwidth extension (ABE) in narrowband telephonic communication. In this regard, we use signal model and H∞optimization to obtain a synthesis filter for representing the wideband information of a signal. We need to estimate the high-band information in narrowband communication. Hence, we construct a high-band filter which retains the high-band information of the synthesis filter. Signal models may not be the same for different speech signals because of their non-stationary (time-varying) behavior. Hence, a short time processing (framing) is applied to speech signals for converting them into the stationary frames. Signal models of stationary frames may be different. As a result, their high-band filters will vary. So, a Gaussian mixture modelling (GMM) codebook approach is used to store the high-band filters information along with their narrowband information (narrowband feature). This approach is also used to estimate the high-band filter information for a given narrowband feature of the signal. Performance analysis is done for the two types of narrowband information representations.