Here's another demo of my synSinger program singing the first verse of "Wildwood Flower", a traditional folk song. (I'm not sure people could take several verses of the synthetic voice ).

Wildwood Flower (synSinger)

The lyrics were entered into BiaB, and the Sound.XML file that was generated for Sinsy was used to create a .wav file, which was loaded back into BiaB.

Backing tracks are the Country Brent Solo Guitar and an added String Quartet track. It's all done in BiaB without any real correction, so it can be heard warts and all.

I've also posted an update to my ever-expanding examples of synSinger singing "Twinkle, Twinkle Little Star":

Twinkle, Twinkle Little Star - synSinger #4

As usual, comments are welcome!

Hi, David !:))


This is very interesting indeed !
I guess that there will come a day
when these virtual singers will
take over (chuckle) but we may
not be exactly there yet ?
In any case, they sing in perfect pitch
which not many of us humans do (arguable ?)
I liked the first one better than twinkle,
and the singer had a very mellow and nice
voice !:))

Cheers
Dani

Thanks for the feedback!

The robotic delivery is certainly a problem I hope to address Real Soon Now(tm).

The first version of synSinger had more "human" features (imperfect pitch, vibrato, etc). But I haven't ported that portion of the code over because there were more pressing issues.

But before I do that, intelligibility is still the biggest problem. I figure only about 40% of the words are understandable, and even that's perhaps a bit too generous. I've just implemented an option for shifting the formants on vowels, and I'll see if that adds some clarity. The voiced plosives need refinement - especially a the end of words, and some of the nasal sounds are of dubious quality.

Hi David,

still very robotic indeed.
But you're on a good way.

Guenter

Hi, Guenter.

Now that I'm over the major coding hump, I hope to be able to focus on getting the little details right.

Although synSinger will eventually be viable, I hope that PG Music seriously considers adding the SoftVoice library to BiaB, since it works well right now.

I'm hoping I can post some demos of SoftVoice in the context of BiaB, but it's complicated by the fact that would require me to see confidential information. That's a bit hard to provide when I'm working on a "competing" product.

The real question for the SoftVoice folk is what degree of realism is required by BiaB users, since any synthetic speech is going to fall short. But I think SoftVoice a strong alternative to Sinsy, and certainly much better than my synSinger currently is (although I'm working hard to bridge that gap).