As I told you elsewhere, I had the same software behavior until November 2023, on another machine with i7. Not anymore now. And I can tell you that I find myself opening quite long complete projects, to change verses or words, because this happens in songwriting.
Careful, talk like that (no matter how truthful it may be) will get you bounced from here.
Ask me how I know.
When I see your video, I can tell you that on my own computer, the difference is awesome.
It takes 10x slower than mine. And it’s a game changer!
OK I agree with you, more accurate processing optimization could be also a need in this kind of software. Especially inside a DAW that requires also huge processing (Multipe effects, Mastering etc.).
Maybe a future version, we could hope.
IA concept is very new and require fast processing (CPU & GPU).
Some would say, earlier than actual technologies… So more power, more possibilities but more cost and consumption.
「いいね！」 1
4th gen i7 is already more than 10 years old and may explain why the render is not very fast for the demanding work of generating a voice.
That said, if you’re using Windows, check if you have the “High Performance” mode set in your “Power options”. This had a big impact on the performance of the computer in my case. Also, note that if you are using a laptop on battery, most operating systems won’t use the full power of your CPU for longer battery life.
I tried your project on my 9th gen i7 and it takes 3 to 5 seconds to be able to preview the changes.
「いいね！」 2
This seems more than plausible
I think I have found this weird issue.
Several times after projects comparisons, I found out that some project more complex haven’t this slower behavior…
So I tried many things, change language, bank, lyrics, notes, etc. nothing was working!
Then I open the projects.svp file and saw a special tag (version): {“version”: 153 etc.
When I update it to 119 (seen in another project) and the issue was gone!
The notes concerned are only updated, not the whole track!
Then I tried a buch of version values and the issue is coming after the version 133 of the svp file.
So I update it to 132 and then it was ok!
Additionally, I compared the files of 2 versions and spotted another tag “instantMode” inside thess json files.
It is set to ‘true’ from version 133.
I tried to set it to ‘false’ (there were many inside the project…) and miracle! The project reacts correctly!
Conclusion, old project could react correctly with huge song but not with last versions!
Tests.zip (673,7 Ko)
「いいね！」 1
ohhh, very interesting. Good work! Hopefully the developers find it, too.
I have a solution for your song.
- select all notes
- set the pitch mode to manual.
- move a note.
- then undo.
- pitch mode manual off again.
well done!
Yes! Another workaround!
Select all notes, pitch mode “manual”, pitch mode to “sing”, that’s all.
Then the render is focused to the closest notes only.
Tag version and instantMode are the same into the JSON file but the render behavior has changed.
A direct way to talk to the SynthV engine, to work differently!
Even on your fast CPU, you get 3 to 5 seconds of choking. Be ware my project is just one verse.
Imagine, when it comes to a full song with standard 3 verses and 3 chorus, you will get six times of choking, which is around half minute before you can preview.
If you constantly change lyrics, adjust note length, add breath, and want to quickly preview, even with your CPU, it is impossible to work with.
I wish your workaround is working, but unfortunately no.
-
Change .svp version to 132 essentially does nothing, except: when you open a version 132 svp file, the software automatically change all “instantMode” from ‘true’ to ‘false’, that what it does. When you save your svp file, the version number is changed back to the new version. In my case, after you save it, the version revert back from 132 to 134, and “instantMode” is remained as ‘false’.
-
Having “instantMode” changed from ‘true’ to ‘false’, is not working. Yes, if you take a look, the live render is only working on the changed part, looking good right? But when you listen to it, you will realize your entire project is ruined. There is no more AI generated sound with natural pitch curve, but everything becomes STD-like sound, and the pitch curve becomes robotic and even.
-
Don’t panic! To fix a ruined project, simply go to the same .svp file, use “find and replace”, to change all “instantMode” from ‘false’ to ‘true’, reopen it, and your project will become live and natural again. However, the live render is back to square one, render the entire track, nothing is fixed.
So, your workaround, is sadly not the way to go.
I tested this method too, there are some issues.
-
It seems you only need to “select all”, “click manual”, “click manual again to turn off”, and that’s it. No need to move/edit a note, and undo, it can still trigger the engine.
-
After you turn off the manual mode, unfortunately, the “pitch deviation parameters” in the bottom of the screen is not automatically reset to zero. You will have leftover curves at the silence gaps.
(see picture) -
If you try to “select all” and delete those leftover curves, the live rendering is going back to square one, will continue to render the full track, nothing is fixed.
-
Everytime you added new notes, you need to repeat the process, having new notes going through the “manual on + manual off” cycle. Can be time consuming while waiting for the manual re-gen and auto regen.
-
If you pay attention to the “green lighting going through the waveform” re-rendering animation, you will realize, although it’s better than regenerating a full track, it’s indeed regenerating a full phrase. You change one note, the full phrase is regenerated. Phrase is defined by silence gaps. So, if your project doesn’t have large enough gaps to separate the sentences, it will regenerate a load of stuff.
It seems Synth V has never had the ability to re-render a few words before and after the modified note. Synth V can only scan for the gaps, and re-render from the beginning to the end of that sentence between two gaps.
In other words, for project using long breath to connect sentences, because Synth V can’t find a gap, it will just render multiple sentences until it detects a gap.
Why can’t Synth V only re-render a few notes before and after the edited one? Doesn’t that make more sense?
I believe the render is focused on all connected notes to the edited one.
If you don’t break the notes with a large enough gap, anything connected will be re-rendered upon change on a single note.
You can test on this example. There are different sizes of gaps to separate sentences.
Project 02.zip (80.4 KB)
I was only pointing out why some projects (or rather tracks) had this behavior and not others. And when I dug around, I found some differences. The anomaly was highlighted to me by the change in behavior of the SynthV engine. Then the subject must be covered by the authors. It took times for me but this issue has no influence on my own projects.
But yes you are right! Updating the project file is only temporary. Each time you save it, for sure, it is back to the last version… So it was not a workaround and I never said that on my previous post.
About your version 134, mine is 153 with the last SynthV version: 1.11.0
I guess the best solution so far, as per your previous post, is to change the svp to version 132, open it, this will trigger the engine, change all “instantMode” to “false”, then you do the editing based on a not-perfect sound. After the editing, save the file, replace all “false” back to “true”, reopen the file. I guess?
Yes it seems to be normal. My previous post was only talking about this opposite “the complete track”, nothing more.
But I understand also if a real singer can sing without breathing, it is a superman!
So, too long sentences may be a bad thing for the singer as well as the listener. But I don’t for the future, maybe…
You can try all the temporary solutions that will allow you to save time. And to work in the best possible conditions, while waiting for a future version which I hope is much better managed from this point. May be also less crashes under Cubase in my case.
Sadly, it’s all about the breath (br note).
It is recommended to connect all sentences with a full length breath (br note) in between, to make transitions natural and smooth.
However, Synth V treats breath as a syllable, not a break, and will live render multiple sentences connected by the br notes, causing unnecessary re-renders.
The only way to improve performance, is to use shorter br notes instead of the full length ones, therefore will generate breaks/gaps.
Sadly, a gap plus a breath, are not as natural as a fully connected br note.
Waiting for the fix in future versions? I wouldn’t bet too much hope for this year.
In the meanwhile, if you all of a sudden, come up with a brilliant idea, then don’t hesitate to post it here. You seem to have a much better understanding as a veteran user than those out-of-reach developers from Tokyo.
I will keep an eye on this post in case any thoughts might strike you in the future.
That’s not what happens. I talked more from my experience than from your example. I have this >5 minute song with 9 voice tracks:
I can change anything in this song and I am still able to preview the result in 3 to 5 seconds. I never had to wait half a minute to preview a change.
You may argue that I am using groups so here is another example where I ungrouped the whole track:
This is from a 3m50 song with 6 voice tracks. When I ungrouped, the whole track started to render, from the begining until the end has you mentionned. However, wherever I position the cursor on those >3 minutes, SynthV starts to render that part in priority. I still get 3 to 5 seconds wait to preview any part of the track, including changes.
If you’re not experiencing something similar, then it looks to me that its more an hardware limitation than a SynthV problem. I suggest you to try using groups or try to optimise your computer settings as I suggested first.
In my opinion, you should evaluate SynthV with groups in your case. You will gain a lot by using groups, not only performance. In my projects, I always end up using groups because it makes it very easy to move verses with the parameters, layer multiple notes on one track and apply changes to all copies of a group (ie. if you have three voices singing the same lyrics, you only change it once).
For me, SynthV works well as it is and my experience is more of the opposite of what you describe. Six months ago, I never thought I could write a song. I’ve written 6 since then and released an EP. I feel it is thanks to SynthV. From my experience, I really do not agree when you say it is “impossible to work with”.
「いいね！」 2