I would like to know what people who have tried her think of her so far. I’ve had her for almost 2 weeks, and honestly, I can’t get myself to use her for anything. She’s incredibly soft and weak. She has no power whatsoever, and the tension parameter doesn’t help at all. The only good thing about her is she has a better tone now, but aside from that, there isn’t much to look for. She kinda sounds like Avanna from Vocaloid now.
I think her low range sounds great. Like I knew before that her old lite version was pitched up but it really shows when compared to the ranges her AI bank is suited to.
I do hope we see a powerful voice in the future but Eleanor was never meant to be one so I don’t mind. That said, she is more proof that unless an AI bank is intentionally designed to be a powerful voice it’s very hard to use it that way.
I suppose I feel the same about Eleanor as I do with the updated versions of Saki AI. The new tech is impressive and sounds great but, because it mimics the original voice provider’s style, AI offers much less flexibility than Standard voices do.
I’ve got mixed feelings. The tone is very good, and she’s very appealing. The nasality is brought down, which was something that really bugged me about the voice.
But the detail isn’t there - she’s lower in resolution both in spectral and temporal detail, and - most problematically - you can hear it. It’s obvious when you compare a spectrogram what’s been lost.
And regenerating an AI voice is much slower, which means every step of the workflow is slower. Moving a note means waiting. Changing a lyric? More waiting. Change the voice parameters? A lot more waiting.
I’ve also encountered more issues, like when SynthV shows that it’s regenerated the song, but refuses to play it back, requiring I close the program and reopen it.
When I have a lot of changes to make, I find that I’ve gone back to the Lite version to make the changes, because it’s so much faster.
As others have said, the power in the voice isn’t really there. I’ve only done a couple of songs with the AI voice, so I haven’t had time to do that much experimentation. With the old voice, I had figured how to adjust Tension, Gender and Breathiness to give the impression of a more powerful voice.
This doesn’t work the quite the same with the AI voice, and when I’ve tried making tweaks to the parameters, it’s so slow to regenerate that it discourages me from experimenting.
I’ve never run across a company that has so many features and virtually no documentation. It’s mind blowing that I have to experiment with things to figure them out, because there’s no manual provided.
For example, I appreciate the new dialog for humanizing the vocal, but what’s the difference between Singing Style - Vibrato 1 and Singing Style - Vibrato 2? What is Tone Shift?
Where’s the documentation? This is really basic stuff that Dreamtonics is stumbling at.
I haven’t formed a final opinion yet, but despite sounding better in some regards, the AI is also a step backwards in a number of ways that make it discouraging to work with.
I’m a fairly new user of Synth V Studio Pro, and this forum. So excuse the lack of a language tag
lang:en as I’ve spent about 30 minutes on and off trying to find where to set this tag.
I had pre-ordered Eleanor Forte AI on the strength of the unbelievable results I was getting with the lite version.
After I got the AI version, and did a straight swap with the lite voicebank, I couldn’t believe what I was hearing. I thought (really really hoped) that tweaking the parameters would help getting it to sound like the lite version. Tweaks would help here and there but, overall, it would always sound like an inferior version of lite. AI just didn’t sound right, to me anyway, and I echo the comments about a lack of power/expression.
I think I should have got the standard version. I know I’m rarely, if ever, going to be using AI for anything other than reminding myself of why I don’t use it.
There is no full Standard version of Eleanor Forte. The reason she was released as an AI bank is because Standard voices need to be recorded on-site, and travel restrictions currently make that impossible.
Volor held a survey about whether people still wanted an AI version even though it would differ from her lite version, since it was the only way to give Eleanor a new voicebank during the pandemic.
I know what you mean, I was disappointed as well when I first tried her. I thought maybe you had to tweak the parameters a bit to make her sound better, but no, she sounds weak no matter what you do. At the moment, I’ve only used her in 2 songs, and only for backing vocals. Don’t see any other use for her. Eleanor Forte Lite remains the best.
I am so new to the voice that I wanted to know quickly if Eleanor could sing. Pop. Over an Orchestra or do rock. She seems to my ears to be able to do all three. I’m not convinced by my music writing but I am convinced the AI voice can work on these song types. Here is a link so you can listen to my test trials and let me know if the standard version would have been better ? Love Needs Fun ( Eleanor Forte AI in three different music styles.) - YouTube
Again, there is no Standard version of Eleanor Forte. There is only the lite version (similar to a Standard voice but with much fewer recorded samples, and cannot be used for commercial purposes) and the AI version.
Thanks . I think I get it. I downloaded a free Eleanor voice into the Synth V pro version to test the software. It sounded poor to my ears compared to the AI version that I finally installed on the 6th December. I’ve found it all very confusing. I purchased the synth V and the voice at the same time in pure hope that they were meant to go together. In this thread I got the comment that the AI version was not as usable as the original. Hope that puts some context on my comments.
PS - what is a lite version. Is that the same as the free version I downloaded ? So sorry if that’s a dumb question.
Yes, the lite version is the free voice. Many lite voices are available at this link: index - powered by h5ai v0.29.2 (https://larsjung.de/h5ai/)
Currently Eleanor Forte and Tsurumaki Maki are the only English voices with lite versions (note that Maki is bilingual and also has a Japanese lite version). Another English lite voice (Anri) will be available in the near future.
Many of the comments above about Eleanor AI’s usability/flexibility are based on previous experiences with vocal synths. AI is a relatively new approach to vocal synthesis, and compared to the traditional concatenation method (ie “Standard” voices) there are some tradeoffs. Eleanor Forte AI is our first opportunity to see the technology in action with a native English-speaking voice provider (since Tsurumaki Maki’s voice provider speaks Japanese natively).
AI mimics the original voice provider’s singing style, whereas Standard is purely composed of detached samples for each phoneme combination, which are then spliced together by the synthesis engine. In my opinion, AI brings much more realism to the voice but also presents the following drawbacks:
- Because the AI is trained on the voice provider’s singing style, the resulting product can struggle to fit other genres or styles. Standard voices are influenced much more by the vocal tone of the provider and tend not to be tied to a specific singing style.
- AI does not have “pitch groups”, which means if the AI algorithm thinks a high note should be falsetto, it is, and the end user doesn’t have much of a say in that. With Standard voices, so long as there are sufficient samples recorded, you can choose between falsetto and powerful high notes.
The natural conclusion is that in order to get a powerful AI voice, the AI must be explicitly and exclusively trained on powerful (“belting”) vocal samples. In short, if an AI voice is not marketed as “powerful”, nobody should expect it to be.
Based on the above, the release of Eleanor Forte AI only cemented my existing expectations. The technology is impressive, but somewhat more difficult to use outside its intended musical aesthetic. I am not disappointed by the product, however as a hobby-level producer I do find it more difficult to achieve my desired sound because it feels like I’m fighting with the AI. Perhaps this is my fault for straying from the vocal style it was trained on, or a result of my inexperience, but that’s been my feeling so far.
lite versions serve as a free trial/demo with these restrictions:
Edit: I see claire wrote a better response already.
“Lite” and “free” are synonymous. If you didn’t pay for the voice, it’s a “Lite” version. In general, “Lite” means that the voice was recorded over a smaller pitch range.
“Standard” and “AI” are different technologies, and there are often “Lite” versions of each. Some voicebanks have both “Standard” and “AI” versions.
“Standard” is generally higher quality in terms of time and frequency resolution, and more responsive to parameters.
“AI” generally means that it’s more convincingly human, but slightly lower quality in terms of time and frequency resolution. They’re a bit more “buzzy”. It’s harder to change from the character of the voice.
For example, most English AI voices tend to be soft, and so it’s harder to make their voices sound “powerful”.
As a note, although Auto Pitch Tuning is only available for AI voices, you can still apply it to a Standard voice.
When you apply Auto Pitch Tuning, it writes Pitch Deviation to the voice’s Parameter track. But if you change the voice, the Parameter track data is retained.
So if you want to use Auto Pitch Tuning with a Standard voice, switch to an AI voice, apply Auto Pitch Tuning, and then change it back to a Standard voice.
Of course, not all Auto Pitch Tuning is the same, so this isn’t guaranteed to give good results by any means. But it can be useful to experiment with.
Massive thanks claire and dcuny. That all makes so much sense now. I tend not to want belting vocals ( possibly old age ) so the AI seemed so much better for me. I do however understand now. The lite version you can tune like a synthesizer to give you almost any tone power and texture you want. Cant thank you enough for explaining. Steven
Couldn’t be happier! Shes capable of everything I did with lite and more. She’s so smooth. I’ll definitely be using them together like I do with other banks like Tsuina STD and Tsuina AI. I love that both of her can work solo or together so well. For me she’s the best example of that. I’m still super looking forward to a full Standard, if they ever get to make one. My favorite vocals all come from Volor, and with this Eleanor being so good I trust the results will be awesome.
I’m very impressed that she is still her recognisable self despite the differences that can occur between recording styles for the different synthesis, let alone the time that’s passed since she was first made! My voice changed a lot in that time haha.
She’s pretty easily pushed beyond her strength in her falsetto range by transposing and saturating her a tiny bit too, I love that she’s so moldable without compromising her quality much. Doing this to Rikka Ai or Maki English AI makes it immediately noticeable. Eleanor can really crank the tone shift and tension either way and maintain intact too!
The new smoothness for pronunciation makes it particularly easy for a non-American to use her without accidentally butchering her accent, too! This has always terrified me a bit with her lite. (shes also what pushed me to finally start composing in the VST in my DAW, she makes it a really pleasant experience!)
One thing I don’t like (which isn’t anything to do with Eleanor really), is that standard banks are slower than AI so using them together can be a little prone to crashing if I move too quickly with the standard lite.
This is very good to know! I hadn’t tried it yet because I hadn’t had much success in the past with Saki AI (not that I tried very hard, because in that case I had the option of simply switching the the Std version).
I’m using your ideas on transposing and saturation for sure. I’ve found the AE voice much better so glad you have too. Steven
Thanks. Then I’ll at least have something to look forward to. Hopefully, things will improve before they run out of variant names.
For a moment there, I thought I was becoming someone that was afraid of change. It’ll be backing vocals for me too, if ever I need them.