Symphonic Choirs Sample

Dorico
#22

I am impressed with the progress that has been made with reproducing choral singing and WordBuilder is a significant advance. I congratulate you, John, with how well you have mastered this complex system.

It reminds me of the time 40 years ago when I was working to get a Commodore Pet to play a simple keyboard piece through a homemade DAC. The programming in assembly language was tedious. It was more computer science than music. I get the same feeling here that what is required is a mastery of sound engineering.

As an musician, I want to spend my time writing. I would love to have an easy way to produce sample choral recordings but I don’t think we are quite there yet.

The central problem seems to be getting the computer to correctly pronounce a notoriously inconsistent language - how do you pronounce knife? - by interpreting the written language. Why has no-one thought to build software based on input from sound rather than writing, to utilize the advances in speech recognition? If you sang the text on a monotone in rhythm, a sound analysis module should be able to select the best matches from the sound library and recreate it with the correct pitches from Dorico as choral singing. I may be totally off base here but I suspect this approach could avoid most of the machinations at the granular level required to get a realistic sounding result.

Thanks for sharing,
David

1 Like
#23

Wordbuilder uses its own phonetic language, Votox. It can automatically translate written English into Votox but the idea is to be able to reproduce any language and with things like Russian and German which I’ve used, it is up to the user to programme the phonetics correctly. Although there are clear gaps in the sort of vowel sounds it can deal with (I do not get the impression it has been programmed by a linguist), it is nonetheless quite powerful.

#24

Neither do I. The repertoire of possible phonemes (vowels and consonants) in EW Symphonic Choirs is obviously aimed at speakers of American English only, who think that’s all the sounds there are, and the rest of the world is doing it wrong :roll_eyes:. I don’t want to even begin to try and reproduce any other language with this tool, despite there being a ‘Latin’ way of encoding stuff. And TBH, I’m not that impressed by its rendition of English either. It sounds like a choir that doesn’t master even basic English (so maybe it would even sound better with a Latin text after all?). Speech synthesis is so much more advanced nowadays.
First step could be implementing a more comprehensive subset of IPA, and supporting IPA notation to begin with. This means users would be required to have knowledge of phonetics. And, yeah, typing IPA would definitely pose a challenge. One can dream…

Then you might also think of the horrific and ubiquitous auto-tune. Pronounce the text in the worst foreign accent or dialect of your choice, and let the computer sing the right pitch. Problem solved :laughing:
@DSale is right, speech recognition is so advanced nowadays, you could leverage it much more in this area. Work in progress, no doubt.

#25

For what it is worth, here’s a demo I did a few years ago in Cubase, with the last part being sung by EW Symphonic Choirs.

https://soundcloud.com/benjischaub/lart-vivant#t=4:28

The Votox was a PITA to get right, but in Latin it sounded okay enough. A major, major issue was the inability of the Wordbuilder to sing the correct lyrics when starting in the middle of a phrase.

Hope it helps!
Cheers, Benji

#26

I like the music in Benji’s example particularly and The Lamb is also a decent example of using the choir and a piece I’ve also sung.

My first two Dorico projects both used Symphonic Choirs but I think a better example would be this Box

It’s a complete if reasonably short Mass written originally in Sibelius. Although I have created a Dorico version as proof of concept, I’ve yet to put in the hours required to balance the dynamics so it’s better to listen to the original audio export (although of course it is itself far from perfect). If 24’ is too long then just sample bits here and there.