I’ve written about my scepticism about AI voices up to recently because most seemed to be pop music orientated and there were no competent choirs. This all changed just over a week ago with SynthV’s release of three choirs simultaneously. This seems to me to be the biggest breakthrough in AI synthesis based on original sampled voices. But they are not just choirs – each one uses 16 separate solo voices which be used in any combination of up to 16 voices per part. Not only is the SynthV software very sophisticated in controlling these but the voices themselves are often surprisingly beautiful as I hope to be able to demonstrate. I’m starting off with a piece I posted previously just before Christmas – the “Noch bist du da” poem. Following the same link latest works now leads to the new version with a solo alto from choir 2. Of course this poem is in German but it’s possible to get a perfectly adequate south German/Austrian accent using the Spanish language model. Of course ideally there should be a German one but I’m willing to wait – French is rumoured to be not far off. Spanish works well for Latin and the words which are clearly wrong can be added to the phoneme dictionary to greatly speed up the editing.
Very nice.
Could you be so kind as to describe your workflow here? I’m guessing something like: Dorico → XML → SynthV (→ DAW?). Then all further tweaking inside SynthV.
This feels like a new milestone in the development of sampled voices, singing words. Beautiful piece, by the way, and also beautiful performance.
appreciate the comments, both of you – many thanks!
yes, workflow first of all Dorcio MusicXML export to SynthV. In the resulting new project, you need to find a way to stop the keyboard acoustic feedback on the SynthV notes – to me the quickest is to simply disable the “natural” p.t with the default Expression Map which is used by anything within SynthV.
Then it’s a matter of choosing the appropriate voice from 8 different altos (12 if you count the vol 1 but that doesn’t really work here) and configuring it for the sound I want. Then the few gradual tempo changes have to be sorted out – generally by just taking the mid point as that’s adequate here (I understand you may be able to do something in Cubase with ARA but that’s not necessary in a work like this).
Next of course is sorting out the pronunciation of the German. If specific words are frequently repeated, you can simply create a new Spanish to German phoneme dictionary and add them — this will be a huge time saver in Latin sacred texts as I already discovered.
Then come dynamics and in places the “mouth opening” parameter set to match. Finally audio export from Dorico.
One workflow thing I discovered is that it’s highly advisable to minimise the SynthV window when initiating playback from Dorico, otherwise both playbacks are active and sometimes you cannot stop playback from either. Could just be my system, I guess but I invite feedback from others on this one. Otherwise the VST plug-in works very nicely. .
Thanks for taking the time.
Yes, mouse clicking the Stop button on both is the only thing that seems to work.
How much time would you estimate it took you to render this piece, start to finish?
probably around 90 minutes – not much more at any rate and I expect as I get more experienced, I’ll get faster. There isn’t a huge amount of text in this one but there were one or two tempi to be sorted out and of course there are always the dynamics.
Interesting ! I experimented Synth V this week but found that MXML export does not include tempo marking and tempo variations (ex. rallentendo), so it required me to tweak the tempo track in Synth V : is there something I am missing ? Moreover, each time I modify my score, exporting the MXLM and importing it in Synth V it creates a new track. Moving the data of the new created track to the track where my voices are set has the consequence to clear all the settings (language, SATB, etc.) : again, is there a way to reimport the MXLM on the exiting track in Synth V without creating a new track ? I agree, Synth V is really promising but as of today it slow down my workflow because of this. Any clues ?
No, it’s either a new track or trash the existing project as a whole. I find that decoupling Synth V from the Dorico workflow is essential. Before I even add Synth V to the VST rack, my Dorico vocal part(s) are 99.9% complete. Once I import the MIDI (or XML) into Synth V, they’re separate projects in my mind and if I must change something in the score, I’ll duplicate the change (if warranted) in Synth V.
EDIT: I should add this is exclusively for creating practice tracks of for my wife’s choir.
Thank you very much for the answer and the advice.
well I find it does include all tempo markings and for the tempo variations it does put in something, though it’s inaccurate and I need to recalculate. Basically I take something just under 50% of the average value of the start and end point of the gradual tempo change and I’m finding it doesn’t take too long to get something accurate, although of course it would be much better if the MusicXML standard did deal with gradual tempi changes properly. However I can imagine that if a score is full of rubato, it could be a real pain with SynthV in Dorico.
Thank you for your insights : I will check again ! Maybe there is an issue with my MXLM export of my parts, because in my case even the first tempo mark is ignored. I wish you a great Sunday evening !
John Barron gave a good tip concerning time management in SynthV in Dorico.
- Make a midi file and a XML file.
- Import first the midi file in SynthV.
- Delete the channels (not the time signs, in midi they are accurate).
- Import the XML file.
John has made a very good video in which he explains how to use SynthV in Dorico. You can look here: Make Dorico sing
I already saw that. When I tried to import the MIDI file, it gave me nothing that isn’t already in the MusicXML – possibly I did something wrong or the latest version does things differently.
Wow, you did a great job with the vocal 
. Do you have all three choirs, or just choir 2? I am trying out choir 2 at the moment, and wonder if it makes sense to purchase all 3, since the offer is quite attractive at the moment.
Thank you for posting this, very inspirational!
Best, Robert
I couldn’t decide between choir 2 and 3. 2 is more smooth and classical but 3 with a slightly more folksong character can be more expressive. I found I needed both so bought the bundle. Choir 1 has limited use in a more classical environment but it’s free after all but it’s well done also.
Anyway, am delighted if I inspire you to get more into this sort of thing