I’m guessing this a holy grail problem, not least because every language and dialect has its own phonetics. Ooh’s and aah’s are often different even within the same language, which is the bedrock of much humour here in the UK!
2 Likes
Right you are. I have dabbled in Yamaha’s VOCALOID (solo singer) software. Even their “English” singing modules have certain vowels that sound Asian. The “oo” sound, for example, sounds more like a French “u” (as in tu) than the “oo” in boot or you. Admittedly, VOCALOID is a relatively inexpensive program compared to others (and not ready for most purposes), but it shows how difficult this area is to conquer.
1 Like
I’m going to be That Guy and say “I don’t believe it’s prohibitively difficult.”
I watched the Alpha-Go documentary last night with my kids. Modern computing is amazing. Go has more possible game combinations than there are atoms in the observable universe, but the computer conquered the game anyways.
The problem isn’t that a realistic wordbuilder is impossible, but that there isn’t perceived to be enough of a market to justify the (admittedly probably significant) development cost. I just think if one came along that were even incrementally more realistic and easier to use, I and others would be lining up in droves to buy it.
Maybe Covid will aid the push towards more realistic virtual tools for choral playback.
3 Likes
I agree.
So do I. When you consider how realistic synthesized speech is for virtual assistants now (Siri, for instance,) the only piece of the puzzle left is modulating the pitch. I think the approach would have to be one of machine learning and artificial synthesis rather than true sample library though. (Or prerecorded pitch sustains with synthesized consonants.)
But it has taken the likes of Google and Apple the prize of millions of users to try and develop the technology. If they ever start to license it, then perhaps we’d se something happen.
Hmm. Siri and her ilk are very noticeably ‘not human’ – and I’d suggest that the problem is in phrasing and intonation, because they just build sentences from components, rather than considering the arc of the phrase.
In singing, these matters are even more crucial, on top of the added difficulty of pitch.
Yes, but now we begin splitting hairs. Siri is darn good compared to days of yore, and perfectly acceptable. So if we are thinking in terms of, better is better, then a Siri-esque singing VST would be a huge step forward compared to the current EW offering. I agree with you that a simulacrum is never as good as the real thing (digital vs. real pipe organ wars… duck and cover!) but digitals are better than nothing, and I suspect we’d be more than happy to have a singing siri in the interim until such time as we have access to real choirs to record our works.
But in the end, you’re right: it is not the same as real humans singing and trained ears would know the difference. I’m just arguing that I think, like Dan, that it could be done and that it would represent a step forward as far as artificial renderings are concerned.
1 Like
Ask Siri to sing a song and she’ll tell you she doesn’t have the brain for it.
1 Like
The answer Siri gave me when I asked her to sing me a song, was that it was more difficult than she believed!
Paolo
1 Like
Here’s my custom choir sound, which gives the slight impression of a changing vowel on each note, and gives the effect of a choral noise, without actually doing any syllables.
For me, it’s good enough as a mock up, to get the idea of what a choir might sound like. It works well against instruments, too, which is often a problem.
1 Like
Me too. I’ve been getting good results with Voices of Prague. Hard work and a horrible and slow UI but good results.
Here’s an update on some vocal samples. Same piece, performed by different libraries.
EW Symphonic Choirs
The tuning is a bit off, and I don’t like the ‘swirling’ you get on long notes. For some reason, a note cuts out in the third chord.
Perhaps there are settings in Opus that you can change to improve things. If you want to use WordBuilder, I dare say there’s some value in having it.
Noteperformer
Like a bunch of slightly unrealistic Oxbridge choral scholars. But you can’t knock the phrasing.
Olympus ‘Sustain Aah’
Woeful tuning, and a bit remote.
OT Miroire Baroque Choir
I quite like this. They have patches for Sustain and ‘Random Syllables’: I’ve mixed in some of the random syllables on short notes in the ex map, just for a bit of fun. “Defaults have been tweaked.”
GPO5 Full Choir KS
A bit slow in the attack, but surprisingly effective. Not great on short notes. Editing the instrument definitions, or making a complex Ex map might improve things.
Benwiggy’s special
GPO5 with some voices from Logic mixed in, to give some oomph. If I didn’t have this, I’d probably go with Miroire out of the box, or be looking for yet another library…
I’ll probably leave it 5 years for the realistic and easy singing of syllables.
6 Likes
there are countless libraries that can sing vowels - many better than any of these examples. But the situation for singing text is still same as two years ago. I agree that in around 5 years we might have a first class option. It seems Arne Wallander is interested in it there’s any way lyrics (which is actually what real singing is all about other than background noise in films) could be read direct from notation software. If he could do something to improve, for instance, the EWQL Wordbuilder, we’d really be talking.
I love those harmonies!
Good ol’ Bill Byrd. (Appropriate for the 400th anniversary of his death!) The first phrase of that “Ave verum” is my favorite illustration of the comma problem in just intonation.
1 Like
Do not forget the Best Service Libraries “Cantus” and “Mystica”.
3 Likes
Mozart: “Ave Verum Corpus”, produced with “Cantus” and “Mystica” (created by Eduardo Tarilonte).Strings recorded with Spitfire BBC Core. Both choir libraries containig a lot of latein phrases, which you can use in a wordbuilder menu.
In fact, the technology for AI singers to perform songs based on scores (usually MIDI) and lyrics already exists. Maybe you can search for ACE Studio and SynthesizerV on YouTube to watch related video demonstrations. They offer a much higher degree of naturalness compared to sample libraries.
ACE studio and Synthesizer 5 are, as far as I’m, aware, still exclusively for pop music. They don’t help for choral music or anything in a more classical style. . For that, the only synthesis engine I know of which is useful is the impressive Cantamus but that works exclusively through uploading MusicXML to their website – you have no control over the end result. For the sort of music I (and probably most Doricians) write, EWQL Hollywood or Symphonic Choirs are still the only serious option as a VST instrument if you want proper text control. There are , of course, libraries with selected phrases (mostly Latin) but that’s no substitute for full control in any language you want.
2 Likes
In fact, I am one of the co-founders of ACE Studio. Recently, I’ve been making an effort to learn Dorico because I want to understand how ACE Studio can better assist composers in creating choral music.
I think ACE Studio currently has a few issues: first, there are not many suitable vocal tones for choir singing (only the opera singer Karolyn and classical singer Barber are relatively suitable); second, even when multiple vocal tracks are layered, it lacks a sense of space and fails to create a choral effect; third, it does not support MusicXML import and the current ACE Studio Bridge VST plugin is not very user-friendly.
We plan to release a new version within a month to address these issues. However, we don’t have team members who specialize in composing choral music, so we are learning as we go and seeking feedback and suggestions from others. I have also tried Cantamus, which you mentioned. Our goal is to achieve the vocal quality of Cantamus while retaining enough editability. Hopefully, in a month, we will be able to deliver a result that I can proudly recommend to everyone!
9 Likes