Reconstruction of wavelet coefficients into separate signals
2 ビュー (過去 30 日間)
I've deconstructed a short audio .wav signal into its 1D wavelet coefficients using wavemenu (db4, 6 levels). This gives mat files ([64*109340 double coefs], [1*64 double] scales, and wname).
I now want to reconstruct each of the D1 to D6 frequency/time scales into separate audio files with the frequency range changed (e.g. to 250Hz centre). I can combine these into one audio signal which has all of the original signal characteristics compressed into a particular frequency range through mixing but it would be instructive and interesting to be able to do this as part of the reconstruction process.
I'm presuming that I should use the wrcoef function but am unsure of which (C,L) coefficients to specify and how to address the frequency range translation issue. I thought that I could fiddle with the reconstruction parameters to achieve this latter aspect.
Any thoughts or advice?
回答 (2 件)
James Russell 2015 年 7 月 9 日
The wavemenu GUI has this capability built-in, good for exploratory purposes. Pick Wavelet Packet 1-D from wavemenu's GUI.Get a signal loaded and analyzed by whatever wavelet, depth, etc. you are interested in. At the right of the Wavelet Packet GUI you'll see buttons for Analyze (which you'll have just used), Compress, De-noise, etc. Below these big buttons, you'll see some labeled buttons: Node Label, Node Action and some others. The default Node Action is Visualize. If you click on any node in the tree with this set, you'll see its coefficients graphed. These are hard (for me anyway) to interpret. If you click on the double-headed arrow by "Visualize" and pick instead "Recons.", you'll see graphed what the signal would look like if you reconstructed it (back to initial sample rate and duration) based on solely that node. This, I find, is much more intuitive.