Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Neural text to speech with dozens of celebrity voices

vocodes.com

Post reply on HN
Print view View original

101–110 of 174 posts

Re: Show HN: Neural text to speech with dozens of celebrity voices

#101

Post by azinman2 »

searchableguy wrote: post #100
azinman2 wrote: post #98

Earlier quoted context omitted.

That’s a big start. If you look at Reddit, you’ll notice all these deepfake porns are using scripts/apps that people have packaged together. Those and any commercial variants can abide by such restrictions which will help document the fakeness once this goes mainstream. Only the super technical would be able to get around it, and if tutorials etc come out then you have legal grounds to go after it to minimize. Don’t…

I don't agree with the solution unless we as a society stop putting trust on any digital media but at that point, this is not necessary. Many governments would love to use this tech so they have incentive to stop others from using it and still let people believe in digital evidence by putting half assed solution like the one you proposed. The cat is out of the bag. Digital media should not be trusted blindly.

I don’t appreciate the condescension. It’s half assed in that I put a rough direction out there for conversation for a huge problem that will inevitably cause social problems. Your reply isn’t in line with HN guidelines and certainly doesn’t make me want to participate in conversation with you about important topics.

Re: Show HN: Neural text to speech with dozens of celebrity voices

#102

Post by slickQ »

tdeck wrote: post #95
afarrell wrote: post #72

Is this ethical?

I too expected more discussion of this. People play around with these things because they're interesting, then mostly hand wave away concerns about the implications with "well, people will just have to learn to be skeptical of recordings". But what we're really doing is muddying a previously reliable avenue of gaining quality evidence about the world. I expect this opinion is unpopular on HN but I think people should…

As NN models get more advanced generating speech synthesis will get progressively more convincing and less expensive to implement, even if the models aren't built for speech synthesis specifically. The same can be said for image generation/transformation. If we are to continue develop AI then this is likely inevitable. There are benefits to these models for mute people, for example. Adversarial models can be built to detect fake audio samples. Regulation (ex: adding tells/signatures in commercial products) would also help. The government would have to ban most AI research or they would only be prolonging the inevitable.

Re: Show HN: Neural text to speech with dozens of celebrity voices

#103

Post by echelon »

tmoney1818 wrote: post #80
echelon wrote: post #28

Earlier quoted context omitted.

I'm not a lawyer, but I think we're entering into a legal gray area. There are the existing frameworks of copyright, parody, free speech, slander, libel, etc. that are all somewhat tangential to this. I believe (I'm not certain) that celebrity voice impersonation is legal as long as it is not used to sell or endorse a product. Most models are trained on the original speaker's voice, but maybe only a little bit. Model…

>"Most models are trained on the original speaker's voice, but maybe only a little bit." Really cool that you got this to work. I used to work on TTS (a few years ago, now), and we trained on celebrity voices, but used full audiobooks. https://github.com/Kyubyong/tacotron Here are some of our Nick Offerman samples: https://soundcloud.com/kyubyong-park/sets/tacotron_nick_215k .

Hey! I've seen your results! Really fantastic work!

Thanks for making this so open and accessible.

Re: Show HN: Neural text to speech with dozens of celebrity voices

#104

Post by echelon »

vmception wrote: post #21

This could make video games take up so much less space and have much more robust speech, especially from NPCs. Subreddit simulator is pretty convincing conversations, putting that to high quality voices? mannnn, so many good applications. Speaking of which, why don't people just talk about the good applications. You'll get ostracized for speculating more bad things about COVID, but talk about how doomed we potentiall…

> This could make video games take up so much less space and have much more robust speech, especially from NPCs.

Maybe, maybe not. You'll see some of the model sizes I posted in comments above. These are quite large, and adding models for multiple speakers gets quite large. These have to live in memory and probably can't be paged in selectively.

Once we achieve high fidelity multi-speaker embedding models (where multiple speakers are encoded in a singular model), then we'll have something compelling. I imagine the models will become less dense over time as well.

Furthermore, if the models are deterministic, then the designers will know what each line will sound like exactly before it's produced.

Re: Show HN: Neural text to speech with dozens of celebrity voices

#105

Post by echelon »

sunsetMurk wrote: post #15

Very cool, and easy to use! Can you give some more info on how you generated the models? I'm also interested in the tech stack you're using to implement this webapp... Would love some details! ..What's next?

> Can you give some more info on how you generated the models?

glow-tts and melgan, which are somewhat unpopular choices given the proliferation of Tacotron2/Waveglow. I chose these due to their sparsity and speed.

> I'm also interested in the tech stack you're using to implement this webapp... Would love some details!

It's a Rust microservice architecture. There's a proxy layer that decodes the request and sends it to the appropriate backend, and then there's the tts service that is horizontally scaled and is responsible for loading the model pipeline and turning requests into audio.

> ..What's next?

For me? Voice conversion in the near term. This takes microphone input and turns it into the target speaker's voice.

I'm also spending a lot of time on photogrammetry. I have a 3d volumetric webcam system right now that I have much bigger plans for.

Re: Show HN: Neural text to speech with dozens of celebrity voices

#107

Post by jereees »

echelon wrote: post #8

This is my pandemic side project, and I'll be happy to answer any questions about it.

You’ve clearly spent a good deal of your time creating this. Bravo. What steps can I take to find more time to dig in projects of my own? Assuming this is not what you make a living out of.

Re: Show HN: Neural text to speech with dozens of celebrity voices

#109

Post by Baeocystin »

echelon wrote: post #68
dsteinman wrote: post #57

Earlier quoted context omitted.

I was going to mention the same. It would be a childhood dream come true to talk to my computer and have it talk back to me in the TNG computer voice.

That's a fantastic suggestion! I'll get to it!

Semi-serious follow-on question- would your model be able to produce voices like GladOS, which are highly processed, but in a consistent manner? Or are there too many assumptions baked in regarding normal human speech?
Post reply on HN