Advertisement

Broadband Spectral Envelope Estimation

Part of the Lecture Notes in Electrical Engineering book series (LNEE, volume 13)

One of the most demanding challenges in bandwidth extension algorithms based on the source-filter model, introduced in Sect. 2.2, is the estimation of the broadband spectral envelope as already mentioned in Sect. 2.3. All methods presented in this book for the estimation of the broadband spectral envelope are based on the narrowband spectral envelope as an input. Experiments have been conducted using additional scalar features and are described in the respective section [Jax 04b]. The extraction of the narrowband spectral envelope as well as the different methods for estimating the broadband spectral envelope will be matter of this chapter. For the extraction of the narrowband spectral envelope we will introduce a method that has been developed within this work for increased robustness against deviations in the telephone bandpass during operation leading to an improved classification. The methods for estimating the broadband spectral envelope out of the narrowband one incorporate codebooks, neural networks and some linear mapping approaches. Statistically motivated approaches using HMMs and GMMs can be found in [Jax 00, Jax 03b, Jax 03a, Park 00, Qian 04b, Yao 05]. Figure 5.1 shows the filter part of the general block diagram for BWE algorithms as introduced within Fig. 2.8 consisting of the extraction of the narrowband spectral envelope anb(n) and the estimation of the broadband spectral envelope âbb(n). Note that the extraction of scalar features is missing for simplicity. The chapter will close with a short discussion.

Keywords

Training Data Feature Vector Hide Layer Speech Signal Spectral Envelope 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer Science+Business Media, LLC 2008

Personalised recommendations