posted 01 May 2010 07:39            

I have a question re the unvoiced technique: is the opposite (voiced -> only the vowels) also possible? Is it possible to extract them somehow?
And another thing: is it possible to integrate spectral timestretching to it, so that you can stretch only the consonants or only the vowels?

Thanks.
Best,

posted 01 May 2010 23:38            

Tau Editor

posted 03 May 2010 07:09            

A little off subject, but just for clarification. There is a misconception the voiced and unvoiced is the difference between vowels and consonants, but its not quite that simple.

All vowels and voiced but all consonants are not necessarily unvoiced. The sounds of the following letters are normally long voiced.
A E I O U L M N R V W Y Z

The following are normally short voiced.
B D G J

The Following are normally long unvoiced.
F H S (SH)

and the following are normally short unvoiced.
C K P Q T X (CH)

Vowels and normally open mouthed long voiced sounds.

I'm probably stating the obvious here but it is sometimes good to clarify.

Pete

P.S another interesting point is that M and N are basically the same sound but it is the transition to the following vowel that is the give away as to which one it is.

also

G is the voiced version of C or K
B is the voiced version of P
D is the voiced version of T
J is the voiced version of (CH)
Z is the voiced version of S
V is the voiced version of F

posted 03 May 2010 11:45            

thanks for clarifying and reminding me again. I'm aware of this (working at an institute for linguistics i should be) but i forgot or ignored it, because i was focussing on the transition idea. I like this "unvoiced sieve". Would it be possible to do the same thing with
voiced sounds? (that is slowly sieving out the voiced parts)

Best,

posted 04 May 2010 15:43            

For example, say you have a signal with only the unvoiced parts coming through. You could put an amplitude follower or peak detector on it to get the amplitude envelope of the unvoiced parts. Let's call this !Amp (which you could get by feeding your envelope into a SoundToGlobalController).

Then you could use (1 - !Amp) in the Scale field of a Level on your original signal. If !Amp is full-on maximum volume during the unvoiced parts, then you have something that would attenuate the unvoiced parts.

To achieve the controllable attenuation of unvoiced parts, you would add a scale factor to !Amp so you could increase or decrease its value until you get attenuated (rather than completely silenced) unvoiced parts. For example:

1 - (!SieveCtrl * !Amp) clipTo01