Extending the Bandwidth of NarrowBand Speech Using Cepstral Linear Prediction
4. Excitation Signal
We generate the excitation signal by upsampling the original narrowband signal by a factor of 2 and then applying this signal to a fullwave rectifier. Here we do a post high pass filtering to ensure that there is no DC bias in the excitation signal. Normalization of the excitation signal is also done to avoid envelope gain distortions.
The use of a whitening filter is used to ensure that the excitation has a flat spectrum. Figure 2a shows the typical spectrum of an excitation for a frame of voiced speech before application of the whitening filter. From this we note that the excitation signal prior to the whitening filter has a low frequency bias that can lead to high frequency envelope attenuation during the spectral envelope recovery process.
Figure 2a: Spectrum of excitation signal before whitening process
Figure 2b: Spectrum of excitation signal after whitening process
Figure 2b shows the typical spectrum of an excitation for a frame of voiced speech with whitening which results in almost 'flat' uniform spectrum essential for preventing spectral distortions during the reconstruction of speech at the enhancement stage.
Previous Page | Next Page
1 | 2 | 3 | 4 | 5 | 6 | 7 | 8
If you found this page useful, bookmark and share it on:
If you are familiar with RSS feeds, you can also sign up for our free blog feed. Our RSS feed is updated in real-time while our newsletter is updated daily.
