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

Lyrebird – An API to copy the voice of anyone

lyrebird.ai

Post reply on HN
Print view View original

131–140 of 311 posts

Re: Lyrebird – An API to copy the voice of anyone

#131

Post by i336_ »

microtherion wrote: post #127
Infernal wrote: post #114

Earlier quoted context omitted.

In the first clip, I'd say 80% of the soundbites were obviously robot-like, but one or two of the "Obama" quotes were startlingly clear - "The good news is, that they will offer the technology to anyone" - I can't hear anything wrong with that in the first clip at all. If they were all that quality I'd say we'd be easily fooled. As a proof of concept this is pretty big.

The "Obama" material sounded quite good, but reverb can cover up a multitude of sins...

Huh.

That means that recordings of speeches/performances from concert halls are potentially suspicious.

Know of any other instances of reverb covering things up? This is interesting!

Re: Lyrebird – An API to copy the voice of anyone

#132

Post by Sunset »

pbhjpbhj wrote: post #123
richmarr wrote: post #71

Earlier quoted context omitted.

In the UK there's a common law thing called "passing off" that's used to protect unregistered IP from impersonation. It's already used to protect unauthorised abuse of voice actors IP.

I'm pretty sure "passing off" requires the seller to be fraudulently claiming the goods are the goods of someone else, that if you up-front say "the voices used are generated by computer algorithm and do not represent any real person" that a claim of passing off would be rendered moot. Trademark/Copyright can't be disclaimed in this way but Passing Off requires active deception AIUI?

How is this not vulnerable to the "Mickey Mouse animated feature film but drawn by 6th graders"?

Re: Lyrebird – An API to copy the voice of anyone

#133

Post by ionwake »

Gaelan wrote: post #15
afinlayson wrote: post #14

This is how a lot of tech companies make proper text2speech, this was just done using the vast amount of audio that's out there for these people. Soon Trump will use this to state that things he's said are fake news. God help us all.

They claim they only needed one minute of sample audio. We really need to start requiring that all public announcements (news, press releases, etc) are digitally signed and put into the blockchain.

But who would sign it?

Re: Lyrebird – An API to copy the voice of anyone

#135

Post by suddensleep »

Ygg2 wrote: post #80
d--b wrote: post #16

Earlier quoted context omitted.

And Black mirror's "would you like to speak with your dead husband again?"

Forget dead husbands, with this tech, it will be hard to trust anything a politician said. Basically, once they master adding this to video, ANYTHING could be construed against anyone. Want a video of a politician saying "Hitler was right" to cheering masses? Want a video about a president saying it's time to start Nuclear War One? You can make that.

From the "Ethics" section of the Lyrebird site:

"Voice recordings are currently considered as strong pieces of evidence in our societies and in particular in jurisdictions of many countries. Our technology questions the validity of such evidence as it allows to easily manipulate audio recordings. This could potentially have dangerous consequences such as misleading diplomats, fraud and more generally any other problem caused by stealing the identity of someone else.

By releasing our technology publicly and making it available to anyone, we want to ensure that there will be no such risks. We hope that everyone will soon be aware that such technology exists and that copying the voice of someone else is possible. More generally, we want to raise attention about the lack of evidence that audio recordings may represent in the near future."

I'm glad the authors addressed this issue pretty forthrightly, but part of me wishes they'd written a bit more about exactly your point. Whether or not recorded speech will continue to be legally binding evidence, I think it's just as important to point out that many people are normally quite happy to take what they hear as solid evidence, especially when it aligns with their prejudices.

Re: Lyrebird – An API to copy the voice of anyone

#136

Post by paraschopra »

I appreciate the ethics link up there in the menu. Not sure if I noticed it on any other AI startup (or for that matter, any startup). Given how complex the world is becoming due to ever increasing co-dependence with tech, I can see how such pages could become as important as 'pricing' or 'sign up' pages. (The privacy issues with Unroll.me, Uber and a thousand other such services will only accelerate this trend).

Good job, team Lyrebird. My feedback is that while the inclusion of ethics page is great, it could do with more content on your vision and what you will not let your tech be used for. I know others can develop similar tech, but it will be good to read about YOUR ethics.

[Edited for clarity]

Re: Lyrebird – An API to copy the voice of anyone

#137

Post by got2surf »

arghwhat wrote: post #60
qeternity wrote: post #45

While all of these vec2speech type models are impressive, I get the feeling that most of the comments didn't listen to any of the samples. It's still distinctly robotic sounding, probably has quite a bit of garbage output that needs to be filtered manually (as many of these nets often have) and is a far cry from fooling a human.

Most of them are extremely mechanic, to the point where it's almost impossible to understand, but others are actually quite convincing. I think it primarily needs to learn to respect punctuation, and to translate them to a breathing pause that matches the target voice ("President having speech"-style long pauses vs. "Politician having their ass handed to them by journalist on TV"-no-air-needed pauses).

Absolutely - listening through the multiple samples with different intonation from both Obama and Trump, some of the samples are much more realistic, while others come off as robotic.

Maybe it would be possible to train the system to prefer certain intonations in certain cases by rating the realism of the speech in context. It would be interesting to analyzes pauses around words grouped by word2vec! Or choosing a "style" of intonations based on punctuation, parameters like words/minute, etc.

Re: Lyrebird – An API to copy the voice of anyone

#138

Post by i336_ »

jpindar wrote: post #105
veli_joza wrote: post #77

Earlier quoted context omitted.

I missed that one, and it's the best I heard so far. Bit of noise in the background, but the voice itself is very believable. I can imagine in year or two the radio and TV will start using this tech instead of actual speakers. Also, translation and synchronization of any content will be so much cheaper.

Try listening to the NOAA weather radio. They've been using very good TTS for years.

Wow, they are. Very surprised.

This is so good I'm wondering whether this is actually a massive (massive) sound bank. I think it might be a sound bank.

A random radio receiver site I found that doesn't require Flash, tuned to NOAA for Akron OH: http://tunein.com/radio/NOAA-Weather-Radio-1624-s88289/

The list I got the above link from (^F "noaa"): http://tunein.com/radio/Weather-c100001531/

I suspect the warbling I'm hearing is not due to TTS imperfections but 64kbps artifacting.

Re: Lyrebird – An API to copy the voice of anyone

#140

Post by got2surf »

This is exciting! If you look at historic speeches (ie from American Rhetoric http://www.americanrhetoric.com/top100speechesall.html), there are large variations in average characteristics between various styles/contexts (on average, pitch/volume/speed are different for inspirational vs somber speeches, for example). But there are also really large differences in the variation - an inspirational speech may be marked by large swings from quiet, reflective pieces to booming, rousing calls-to-action while a somber speech has fewer swings in delivery.

For the examples given for various intonations from Obama/Trump, some intonations are much more natural than others. It would be interesting to decide how to parametrize a sentence for the intended intonation. (based on word2vec analysis of the words in the sentence, punctuation cues in the sentence, and perhaps a specified category of "emotional delivery").

It would be interesting at the sentence-level, but also at the macro speech-level to include the right "mix" of intonations for a specific context. On a related note, it would be interesting to study the patterns of intonations in successful vs unsuccessful outbound sales calls, for example, to learn how to best simulate a good human sales voice.

Post reply on HN