Sonarworks SoundID VoiceAI - First attempt!

[CrimsonWarlock]

I tried it on a 'fake' female backing from one of my older tracks, which was manageable in the mix but pretty bad on its own, and the result was much better than what I started with.

So I just bought it

 

[HCMarkus]

I just can't see releasing a track using it as the featured vocalist... regardless of how good it may sound. Feels like walking too close to the edge, at least in this point in my experience and personal moral code.

Not to mention the excellent singers I regularly work with wanting to crush my skull with a sledgehammer.

 

[David Cuny]

Here are some more male examples. Can you tell which is human and which is VoiceAI? And more importantly, could a casual listener?

I can't tell, but Example C sounds like it's got some artifacts in the high frequency content.

Other than that, they all sound usable, which is the point, right?

I then applied VoiceAI to her sped up vocal, and voila! The result was significantly better - no weird pronunciation, no strange artifacts. In fact, VoiceAI did a much better job at articulating the lyrics. We were both quite surprised and somewhat impressed.

I wonder if the voices were trained at a particular rate, such as 120 bpm. I wouldn't think that would make much difference, as consonant articulations are - as I understand them - at a fairly fixed rate, and vowel to vowel transitions are pretty broad.

Since this is a neural network, it may have learned some consonant transitions (and thus, indicators of the following consonants) at a faster rate. So speeding up the audio might bring the source material in closer alignment with the training material.

 

[Nekujak]

just can't see releasing a track using it as the featured vocalist... regardless of how good it may sound. Feels like walking too close to the edge, at least in this point in my experience and personal moral code.

I understand the moral conundrum - I'm going through it myself.

But one thing to consider is the nature of the final product. Working on modern pop music, where quite frankly, so much of the genre leans on heavy processing and artificial sounds, not to mention over-use of AutoTune, I see VoiceAI fitting right in, assuming the quality is up to par, of course.

The other consideration is a simple question: how is VoiceAI really any different than the typical sample library? Is using a sampled/modeled voice really that different from using sampled strings? With one, I play the keyboard and strings come out the other end. With the other, I sing a melody and a different voice comes out the other end. Both are transforming my performance into something that's artificial.

So if I'm going to draw a moral boundary, the only logical line for me would be the deliniation between using only real instruments versus using sample libraries, including VoiceAI. If I'm willing to use sample libraries in my productions, then I've already crossed the line so to speak, and tools like VoiceAI are really just another sound source.

Anyway... it's interesting just to be considering these things, regardless of how they end up being used.

 

[HCMarkus]

I understand the moral conundrum - I'm going through it myself.

But one thing to consider is the nature of the final product. Working on modern pop music, where quite frankly, so much of the genre leans on heavy processing and artificial sounds, not to mention over-use of AutoTune, I see VoiceAI fitting right in, assuming the quality is up to par, of course.

The other consideration is a simple question: how is VoiceAI really any different than the typical sample library? Is using a sampled/modeled voice really that different from using sampled strings? With one, I play the keyboard and strings come out the other end. With the other, I sing a melody and a different voice comes out the other end. Both are transforming my performance into something that's artificial.

So if I'm going to draw a moral boundary, the only logical line for me would be the deliniation between using only real instruments versus using sample libraries, including VoiceAI. If I'm willing to use sample libraries in my productions, then I've already crossed the line so to speak, and tools like VoiceAI are really just another sound source.

Anyway... it's interesting just to be considering these things, regardless of how they end up being used.

Definitely interesting. The fact that VoiceAI models the voice that it is processing keeps it closer to "we are actually creating something" vs writing AI prompts.

 

[Nekujak]

Definitely interesting. The fact that VoiceAI models the voice that it is processing keeps it closer to "we are actually creating something" vs writing AI prompts.

Yeah, I definitely don't put VoiceAI in the same category as prompt-driven music generators. There might be AI involved, but VoiceAI is not creating a musical performance out of thin air. It's solely dependent on the performance you provide. In fact, I think it's kind of interesting that in my last post of vocal examples, it mimicked my faux Southern drawl.

 

[CrimsonWarlock]

The fact that VoiceAI models the voice that it is processing keeps it closer to "we are actually creating something" vs writing AI prompts.

That's where I am with this. I want my music to be my performance, and with this I'm still performing it. It is like playing something on a Hammond organ instead of a piano. You are still playing the part, but just choose to use a different 'voice'.

 

[HCMarkus]

You might want to reach out to Sonarworks. If it was this close to the sale, they might give you a refund, or maybe give you a free license to one of the expansion voice packs.

Thanks for that... it lead to this:

Thank you for your purchase and for checking in!

As you made the purchase less than 7 days before the start of the sale, I'm able to match the sales price under our Grace Period. I have issued a refund for the difference and you can expect the funds back via the used payment method.

Enjoy the software and I wish you a nice week!
Kind regards,

Marco from Sonarworks

Nice!

 

[Nekujak]

Thanks for that... it lead to this:

Nice!

Sweet - glad it worked out!

 

[CrimsonWarlock]

Have any of you purchased the rock voices expansion? And if so, can you post some examples?

 

[Nekujak]

Have any of you purchased the rock voices expansion? And if so, can you post some examples?

I only have the pop voices expansion, but I found this YT video that goes through some, if not all, of the rock voices. The video is in French, but the vocal examples are English:

​

 

[CrimsonWarlock]

I only have the pop voices expansion, but I found this YT video that goes through some, if not all, of the rock voices.

Thanks for finding this. Not all the Rock voices were there, but enough to get an idea. Clearly (and probably obviously, for commercial reasons), the rock voices seem very much aimed at progressive metal but are likely manageable for my genre (progressive symphonic rock). Not enough, though, to get them right away (maybe I'll wait for another sale ). I'll see where I get with the default voices first.

Another observation I got from that video is that pitching voices inside the plugin tends to get into Donald Duck territory pretty fast, especially the kid voices. My guess is that you might get better results with pitching the original voice first and tuning the formants before running VoiceAI on it. Haven't tried it myself yet, but surely will look into that.

 

[Nekujak]

Another observation I got from that video is that pitching voices inside the plugin tends to get into Donald Duck territory pretty fast, especially the kid voices. My guess is that you might get better results with pitching the original voice first and tuning the formants before running VoiceAI on it. Haven't tried it myself yet, but surely will look into that.

Yes, I try to get the source vocal sounding as good as possible before running it through VoiceAI, since the plugin is entirely dependent on the original vocal performance. Then after VoiceAI generates its vocal, I'll apply any compression, saturation, and dessing.

But I'm certainly not an expert - this is just the flow that makes intuitive sense to me.

 

[Nekujak]

Just wanted to add that IMHO, the voice expansion packs don't really seem to be that great a deal.

With the original plugin, you get 30 voices (I think it's 31 actually) - basically 15 male and 15 female. Each voice expansion pack gives you 10 voices, which on the surface doesn't seem bad, until you realize that's only 5 voices in each gender. So if you're primarily working with male-only or female-only material, you don't really get a very large selection of options.

On top of that, there are no discounts on the voice packs. So far, the best way to buy them that I know of, is through JRR using the "forum" discount code, which brings them down to just over $25.

With the pop voices, I found them to be very similar to the voices already included with the plugin, but I can see how the rock voices definitely add a substantially different character, so I would think the rock expansion is definitely worth considering.

 

[Nekujak]

Just wanted to post an update on my journey so far with SoundID VoiceAI...

At present, I've finished producing 6 songs of my current 10-song pop project, and up to this point, have decided not to use VoiceAI at all, not even for background vocals. This was partly a functional decision - VoiceAI is not quite good enough yet to carry a convincing lead vocal for an entire song, and isn't really necessary for imparting character and variety to backgrounds; and partly a moral choice - I wanted to avoid having to explain the use of AI in my productions, even if it's in an assistive capacity, so before resorting to any kind of AI help, prefer to use human contributions.

So... I hired a singer to record vocals on songs where my wife's voice wasn't a good fit. Don't worry, this was at my wife's suggestion, so all is still well at home . When needed, I've applied some formant shifting and EQ effects to alter the character of a vocal, but for the most part, have stuck with the singers' natural voices.

During the writing process though, I have used VoiceAI to transform my quick rough vocal sketches into a female voice to hear how it will sound. This has actually been quite helpful in zeroing-in on a sound or style that I'm after, and for creating quick demos that the singers can work from.


I'm actually not opposed to using VoiceAI to transform lead and background vocals in final productions. It's just that for this particular project, given some of its limitations and the current attitudes surrounding AI, I'm choosing to sidestep it altogether. It's possible I might choose to use VoiceAI in the future.

By the way, it looks like Sonarworks has been trying to make their rather ugly VoiceAI web pages slightly more useful. They now have a page where you can audition all the voices they offer, both in the core product and all expansions.

Has anyone else been using VoiceAI, ReSing, Ace Studio, etc. and have anything interesting to report?

 

[David Cuny]

I've had a couple of occasions where I've loaded vocals into SynthV to see how well the worked. Sometimes it's sounded pretty good.

But... It's usually pretty awful at guessing the actual words. But unlike ReSing or Vocoflex you can actually change words that it gets wrong. Just be prepared to do a lot of correcting.

I'm still waiting for all these other tools to have some sort of "correction" feature to make them useful to me.

 

[Mike Greene]

Here are some more male examples. Can you tell which is human and which is VoiceAI? And more importantly, could a casual listener?

VoiceAI Example A:
View attachment VoiceAI-A.mp3

VoiceAI Example B:
View attachment VoiceAI-B.mp3

VoiceAI Example C:
View attachment VoiceAI-C.mp3

VoiceAI Example D:
View attachment VoiceAI-D.mp3

VoiceAI Example E:
View attachment VoiceAI-E.mp3

I think a casual listener would not be able to tell, and even in my own case, they're all "believable."

But ... they're not all good, and I think that's the tell. Hopefully I'm not about to offend, but I would reject C, D and E as not great interpretations of the song. (I'd be telling the singer to try again.) A and B are both good IMO, but it's hard to tell which is the original, because they're so similar in timbre that VoiceAI didn't futz with them as much.

Then again, it it turns out D or E are the original (I'll eat my hat if C is the real vocal), then I'll have to concede VoiceAi improved the interpretation.

If I'm right, though, I think this speaks to the danger with voice-changing, where we have to be careful the "new voice" isn't botching the delivery. Someone with a different timbre is probably going to sing lines differently than the first singer, since singers know to play to their voice's strengths.

An extreme example would be with "Tyler" in the opening post:

SoundID: Tyler
View attachment 05 Carefree Moon - Tyler.mp3

The tone is cool, but it just isn't how a guy with a voice like that would sing this song. He could sound great doing this song, but he'd need to approach the lines differently.

 

[Nekujak]

Then again, it it turns out D or E are the original (I'll eat my hat if C is the real vocal), then I'll have to concede VoiceAi improved the interpretation.

Well sadly, I must admit that D is my own inadequate vocal - but I agree with you 100% (I always hated the way it finished weakly on the word "dead", among other flaws). I never professed to be a singer, even though I've inflicted my talentless voice on many a bewildered audience, and thankfully, never released this particular track commercially

Believe me, I've gottan an earful of critiques from my wife every time I try to sing something, so I've developed a thick skin to the point if someone happens to compliment my singing, I never believe them

Actually, I find it kind of encouraging that VoiceAI was able to impart some modest improvement. It gives me hope that I might actually be able to use VoiceAI to shore up some of my other deficient vocals.

Your assessments are all spot on, including your comments about Tyler's vocal

 

[HCMarkus]

Believe me, I've gottan an earful of critiques from my wife every time I try to sing something, so I've developed a thick skin to the point if someone happens to compliment my singing, I never believe them

Singing a lead vocal may be the ultimate form of musical expression but one's soul risks torture.

Background vocals are safer.

 

[Nekujak]

Singing a lead vocal may be the ultimate form of musical expression but one's soul risks torture.

Background vocals are safer.

Unfortunately, background singers are usually far superior vocalists than the front man/woman they're backing. Take Lisa Fisher, for example. I know for certain I could never do that job .

I'm better off pretending to sing and hope the audience will forgive my ineptitude in the same way they indulge Bob Dylan, Tom Waits, Kris Kristofferson, Leonard Cohen, etc...