Secret Phonemes that aren't documented anywhere?

[unknownlinguist]

There’s a phoneme [pau], in addition to the other “common phonemes”.

Common phonemes: [cl] = glottal stop , [br] = breath , [sil] = silence

[pau] functions in every language, just like any of those 3. If you put it by itself (or surrounded by [sil]) it makes no sound at all. however if you replace [sil] with [cl] or [br], it will make a sound. I think that’s interesting because that indicates a differing functionality from [sil].

“pau” is probably short for pause, right? So I’ve tried using it for words with a pause in the middle, that kind of thing. I think it might be good for putting in place of really short gaps between notes, but I don’t think it sounds any better than using [cl] , [sil], or even a short [br]. Maybe it’s meant for something else?

Anyone find any other mystery phonemes? If I remember correctly, in earlier versions of the program, you could append a number to (seemingly) any phoneme to get an alternate recording of that sound subbed in. I think it only worked on standard voicebanks, but I’m not sure. The only reason I tried it was because I used to use moresampler & arpasing, and wanted to see if the same functionality existed in Synth-V. I found some pretty funny sounds by trial and error a while back.
(I checked and it seems like you can still put a number at the end of every phoneme, but it seems to always sound like silence despite changing the appearance of the waveform below the note as well as appearing as a phoneme would within that waveform graphic)

I just remembered about [pau] today when looking at the wikipedia page for ARPABET, and noticed that PAU is listed in the table for TIMIT there. That made me try inputting some of the other phonemes/stresses/symbols listed on the page, but besides [pau] I haven’t found any more than what is listed here

Anyone find anything else like this?

[claire]

The “secret” phonemes were omitted from Dreamtonics’ website because they’re from early development and can have unpredictable results. It’s likely they were used in early versions of the software but eventually were deemed redundant.

For example, the Japanese kw and gw exist in the phoneme set but are not programmed into the voice databases, so they either produce noise or an unpredictable vowel sound that varies depending on the chosen voice.

Similarly, English q represents a glottal stop in Arpabet, and is used as such for the original Eleanor Forte lite, but has since been dropped in favor of the universal cl phoneme that does the same thing much better. Even with Eleanor Forte lite, the q was poorly programmed and had some odd aspiration at the tail end that you’d have to cut out with the loudness parameter or in your DAW.

Now q behaves like kw and gw as an unpredictable vowel sound, usually somewhere between uh and ow, or just sounds like noise. I’ve also seen mention of English tw and dw, but I’ve never seen those produce output. Words that would use those phonemes like “twine” simply use the two separate phonemes t and w.

The numbered phonemes you’re thinking of were probably the original breath samples. Standard voice databases (specifically the Quadimension Medium5 ones and Saki Standard) had breath samples that could be inserted using br1 or brl1 (breath and long breath respectively). The exact number of variations is different for each voice database. These phoneme codes are obsolete for AI voice databases because there is not a distinct set of breath samples to choose from.

Variations of phonemes can be cycled using the buttons at the bottom of the Note Properties panel: