Sonarworks SoundID VoiceAI

[JohnPativens]

Nice! Does Soundworks require any ID Card for registration like Dreamtonic does for its AI software?

Just for clarity, Dreamtonics Synthesizer V (their better program imho) does not require ID (at least not for me). You're thinking of the Dreamtonics Vocoflex program.

Yes, I meant Vocoflex

[Cuauhtli]

Nice! Does Soundworks require any ID Card for registration like Dreamtonic does for its AI software?

No, but they require a constant internet connection. Are they stealing your data and selling it? Are they just trying to protect their intellectual property? Who knows.

Glad you can install/uninstall voices individually.

[JohnPativens]

Nice! Does Soundworks require any ID Card for registration like Dreamtonic does for its AI software?

No, but they require a constant internet connection. Are they stealing your data and selling it? Are they just trying to protect their intellectual property? Who knows.

Glad you can install/uninstall voices individually.

AFAIK the perpetual license does not require cloud processing, it should work offline also.
Can anybody confirm?

[Cuauhtli]

From their website: Note: While the audio processing takes place on your CPU locally, an internet connection is still required for the license to be recognized and for downloading/updating the database (offline use is not supported).

Hence why I thought you needed a constant connection to use it. But I just tried it and it does work offline.

One thing I have noticed is there is a vocal part where I modulated the pitch. I quite like it because it is actually in tune and it was natural (not autotuned). Anyway VoiceAI can't handle it. It sounds like it's pitch tracking tries to jump an octave instead of a couple of semitones. And it sounds like a computer glitch, not just an octave jump. Like a robot having technical difficulties. Does it with every voice male or female.

[Trancit]

... But I just tried it and it does work offline.

I guess with offline processing it doesn´t need the internet connection always but from time to time to check the licenses...

One thing I have noticed is there is a vocal part where I modulated the pitch. I quite like it because it is actually in tune and it was natural (not autotuned). Anyway VoiceAI can't handle it. It sounds like it's pitch tracking tries to jump an octave instead of a couple of semitones. And it sounds like a computer glitch, not just an octave jump. Like a robot having technical difficulties. Does it with every voice male or female.

I would use the plugin first to transform the singer before doing creative pitch editing ... the more stable the pitch the better recognizeable it is for the processing.
So I would guess, everything drastic like bigger pitch slides can easily confuse the algorhythm in the way it tries to analyze the material... I would apply them afterwards or stay away from samples being already printed that way...
Same goes for effects being applied to the audio like reverb, Chorus etc...
I can imagine the best results you´ll achieve with dry and clean source material

[Cuauhtli]

The pitch slide was part of the original recording. I could fix it in melodyne but now that I've experienced how well this follows the vocal character I'm motivated to re-record the vocals anyway and try a few things.

I'm even going to try running vocoded vocals and chipspeech through it and see what happens. Probably nothing special but still worth a try.

[Cuauhtli]

Well, apparently chipspeech sings better than I do. It sounded as good if not better than my crappy vocals going through Voice AI.

Anyway, this thing has a few bugs. I've had it fail to render the complete vocal file/sample several times. It goes through the whole processing cycle and says it's done. And then only a part is rendered. Other times I've tried rendering a verse at a time and sometimes when it's done, it plays back the wrong verse (one I'd rendered earlier or maybe even from a previous session). Which got me looking for the app's cache. You can open it from the app so it's easy to find. It saves all your renders I guess, so keep that in mind as you may end up with GBs of drive space used up if you forget to clean it out. On the other hand it may save you some time and effort if your DAW was to crash.

Still quite impressed with the quality of the renders. Pitch tracking seems very sensitive. Great in some situations, but also problematic in others. If you like to sing with vibrato you may run into issues. Voice AI tracks well but it doesn't handle pitch modulation very smoothly in my experience (keep in mind I'm a crappy singer). I found turning the "clean up vocals' option helps and some voices handle it better than others. Wish it had a Barry White type of voice included though.

[JohnPativens]

Sounds like you're enjoying it! Did you buy the perpetual license or spare tokens?
I'm still on the fence. At the moment I don't need it but it's useful for sure for demos and such.
I think IKM Resing might grow better in the next year fixing bugs and adding more languages, and if I know their pricing strategy well, it's gonna be periodically discounted other than included in the next Total Studio Max.
They did a great job with the demos, even though someone said that at the moment, in average user experience, VoiceAI sounds better.

[Cuauhtli]

I got the perpetual license and extra voice packs sale they had. I'm surprised how well it works. The rendering isn't fast but I guess that's to be expected. I was going to time it but forgot to. Best practice would be to render a short piece when demoing voices, though you do have to process each time you change voices or pitch. The first time I rendered a female voice I forgot to pitch it up an octave. It sounded like an 80-year-old woman who smoked three cartons a day for 60 of those years.

The main caveat would be the quality of the voices included. I haven't tried them all, but some are definitely better than others tonally/timbre wise. Though, to be fair, I haven't tried adjusting the pitch to match the chosen voice's ideal tonal center, which Voice AI does state in the description. For male voices I leave it at 0 or set it to Auto and for female voices I set it for an octave up.

You definitely should demo before you buy. There's no learning curve to using it. The quality of the voices will probably be the determining factor. The quality of the renders depends on the quality of the original vocal. Personally, I hate the sound of my voice, so anything is an improvement, and music is a hobby not a job, so this is just for personal enjoyment/amusement.

The Resing demos do sound pretty good. And the ability to train your own or download the freely available rvc voices is cool.

[JohnPativens]

Their demo doesn't feat all the vocal models though, do they?

[Cuauhtli]

I think 6 voices and two instruments. At least that's what the installation manager showed when I demo'd it.

Hard to critique the voices. I think it's mostly going to be if you like their tone/timbre and if they fit the original material. There aren't any Whitney Houston or Barry White type of voices that I've found in it. Or whomever is considered a good singer nowadays. I'll use some vocal samples and see how it sounds with a proper singing voice and report back.

Edit: Okay I take back my criticism of it having issues with vibrato/pitch modulation. Using samples of proper singers, it followed the vibrato fairly well with no artificial artifacts (like it did with my vocals), as well as some acapella stuff. Some of the voices do better than others. Maybe this can be improved if you set the pitch to the selected voice model's sweet spot.

But definitely demo first. I'm a lousy singer so it is an improvement for me. Others may find it artificial sounding or the quality of the included voices lacking. It might be good enough for backing/doubling vocals or demo work.

[Cuauhtli]

Just realized something that should probably be mentioned. Maybe it's obvious since it's AI and it's just replacing the tonal footprint (I guess) but it handles foreign languages.

And apparently I sing better in Spanish than English. Go figure.

[JohnPativens]

Sonarworks say the model could apply english accents in other languages if the modeled singer is English. Resing also has English and Spanish models.

[JohnPativens]

I bought a perpetual license in the end.
It works. I hear a strong artificial tuning on soloed processed vocals though. Maybe it's less evident in the mix. I don't hear the same in Resing audio demos, I will try that when the free demo version will be released.

Wouldn't use for lead vocals, for demos and backings seems pretty good anyway.

[Cuauhtli]

It depends on the vocals and voice used maybe. I did have a lot of that on some of my vocals but seemed okay on other vocals and some of the AI voices seem to do better than others (though that may be my imagination).

In any case, there is apparently a 'freemium version' now. I think it includes 5 voices and 5 instruments.