27 comments

[ 4.7 ms ] story [ 268 ms ] thread
All these TTS services look cool but I don’t know how any of them are different from each other…
The price is lower than the competitors... I wonder how good it actually is though. Guessing they just sacrifice quality
the prosody seems a little robotic, and kind of jarring. maybe I'm spoiled by Bark, even in its rough and slow state, but is this really that much of a step up from Tacotron2?
Which is the best one, so far? Eleven labs? OpenAI apparently has something in the works to (going by today's Spotify podcast updates).
Still Azure Speech in my experience.
ElevenLabs is better quality wise, but it's vastly more expensive. Azure Speech hits a really good price:quality ratio.
I'm also curious. A review of what's state-of-the-art today would be a great idea for a blog post. Just don't post it on medium.com please
Google Soundstorm had the best demo so far. It takes few seconds of original audio and continues it with the same voices. Just hearing those examples you wont figure out where original finished and generated one started.
Yeah, neural codecs are pretty amazing. The most incredible part is that they can do compression well across the temporal domain, something which has been non-trivial.
"Ultrarealistic AI Voice Generator" - I initially read the title as "Unrealistic AI Voice Generator". I'd suggest adding a space to "Ultrarealistic".
Every time I see one of those, as a big fan of TV crime dramas, I cannot help but think that voice recordings as proof are going to be a thing in the past very soon.
What are the Ts&Cs when it comes to cloning voices?
Frankly, these don't sound any better than Google Cloud's TTS, and is orders of magnitude more expensive.
I think the killer feature here is supposed to be voice cloning, which IIUC Google Cloud offers only as a custom enterprise thing that takes weeks (which suggests that it's not fully automated).
Cloning is nice, but what's the point if it doesn't sound natural? Will people really pay 100-1000x (if not more) just to get their preferred voice, but won't sound anything like that person when speaking?
It does sounds pretty good though. Plus, I think it’s more like 2-3x, not 1000x
> Plus, I think it’s more like 2-3x, not 1000x

The most expensive Google Cloud option[1] gives 100,000 bytes free per month. This gives 10,000 chars per month.

The next tier is $5/month. With Google that gives you over 130K bytes.

And that's the most expensive model. With the moderately priced model $5 gives you 1.3M bytes.

With the cheapest model you get 5.25M bytes.

I use their service, so I know the economics. I pay nothing per month because I use the intermediate model, and I don't need more than 1M bytes per month. Switching to this, I will have to pay at least $5/month for a voice that's not really any better than Google's.

[1] https://cloud.google.com/text-to-speech/pricing

Interesting but there’s a weird pause before the word ‘danced’ in each demo? Sounds unnatural
Hey everyone - I'm the founder of Fluxon. Just saw that we were on HN today - sorry for the late replies.

The app you're using right now is essentially an alpha-version - even the website was only made this week. Sorry about the somewhat-broken experience so far.

I'll try to get back to everything here in the next couple hours but if I miss something / you have other questions, please ping me on akshay@fluxon.ai

Hey there! Just FYI your APIs are reporting address not found: Error: getaddrinfo ENOTFOUND app.fluxon.dev
Sorry about that! Docs are fixed now.

(Other cofounder of Fluxon)