posted 07 February 2013 15:33            

I have watched the TAU editor videos a couple of times, but my experience is coming out a bit different. Let me explain what I am doing:

1) Take two samples of a note (C3) played on two different but similar physically modelled instruments
2) Go in to the TAU editor
3) Press the + button to add the analyse the sample (once for each sample)
4) The first thing I notice is that when I play back the samples, they are already grainy and have lost a lot of their sonic qualities. I also noticed that there are far too many anchor points to manually merge the envelopes as per the first TAU editor video.
5) I then do a blanket merge of everything using the tool in the drop down list.
6) I am then able to morph the different envelopes with the virtual interface: amplitude, spectral, frequency etc.
7) The result sounds kinda cool - especially when I map the faders to my touch osc controller, but not musical. It's quite cacophonous.

posted 07 February 2013 16:05            

Since you noticed that it is identifying too many segments as unvoiced segments, look at the Bandwidth envelope. When the bandwidth is above a certain threshold, the segment is considered unvoiced. Move the crosshairs to find a horizontal threshold that you think more accurately separates voiced from unvoiced.

Then try reanalyzing by clicking the large button at the top of the TAU where it shows the Range (low to high frequency) and Threshold. Use the new threshold that you have determined by looking at the bandwidth envelope. And give it a range of frequencies that more closely matches what you expect the sample to have.

In the Help menu, there are links to some additional documentation on Tau Analysis and the Batch Analysis Tool.