Re: Real Singer - Virtual Singer
« Reply #15 on: Nov 25th, 2002, 4:51pm »
Quote | Modify
COMAlite has stated something that I also stated recently, using different terminology, on a different VS forum thread.
I do not believe that RS needs to record and analyze unvoiced plosives, aspirants, or fricatives (unvoiced consonant phonemes). It does not seem to me that the vocal quality of one person, compared to another, is distinguished by these phonemes. There are differences in whether or not a consonant is expressed loudly or softly, or elided (omitted). But this can be handled with VS technology.
It is the voiced consonants, and the vowels, that most distinguish the tonality of one person compared to another (in my opinion). I believe that this can be effectively analyzed by a modified version of Real Singer, then "pasted into" VS, which will supply the unvoiced phonemes.
My original suggestion was that VS simply "mimic" the voiced phonemes by auto-adjustment of is own internal parameters (formants).
Related to this, but not as a matter of formant analysis: Another VS thread asked how to extend the sound of "n" in the English word "son". It can be done in VS. As an exercise, I tried singing it in various styles. What I found is that the sung word did have an extended "n", although not as extended as in the posted forum question. But the spoken word "son" does not have the extended "n". Thus, it is not accurate to merely sing the spoken word.
I suggest that the VS voice editing be modified to allow the user to change the relative length of phonemes in a syllable. The present method of adjustment is too complex. A simple slider, with moving parts to delimit each component phoneme in a syllable, would be best. It would be accessed on a syllable-by-syllable basis by double-clicking a note, rather than by the VS control panel.