Artificial Bandwidth Extension Using H∞ Optimization

Deepika Gupta, Hanumant Singh Shekhawat


This work proposes a new method for artificial bandwidth extension (ABE) that aims to extend the bandwidth of speech signals in narrowband voice communications. We extract a signal model which consists of the wideband information. Using the signal model, we obtain an infinite impulse response (IIR) interpolation filter with the help of H∞ optimization. Interpolation filters are going to be distinct for the speech signals because of their non-stationary (time-variant) nature. In narrowband communications, only narrowband signal is accessible. Hence, a codebook approach is intended to keep the IIR interpolation filters information (wideband feature) together with their corresponding narrowband signal characteristic (narrowband attribute). For that, the Gaussian mixture modeling (GMM) codebook approach is utilized to estimate the wideband feature for a given narrowband attribute of the signal. Performances are assessed for the two sorts of narrowband attributes.


 DOI: 10.21437/Interspeech.2019-1580

Cite as: Gupta, D., Shekhawat, H.S. (2019) Artificial Bandwidth Extension Using H∞ Optimization. Proc. Interspeech 2019, 3421-3425, DOI: 10.21437/Interspeech.2019-1580.


@inproceedings{Gupta2019,
  author={Deepika Gupta and Hanumant Singh Shekhawat},
  title={{Artificial Bandwidth Extension Using H∞ Optimization}},
  year=2019,
  booktitle={Proc. Interspeech 2019},
  pages={3421--3425},
  doi={10.21437/Interspeech.2019-1580},
  url={http://dx.doi.org/10.21437/Interspeech.2019-1580}
}