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

Deep Voice: Real-Time Neural Text-To-Speech

research.baidu.com

Post reply on HN
Print view View original

41–50 of 80 posts

Re: Deep Voice: Real-Time Neural Text-To-Speech

#41

Post by visarga »

monk_e_boy wrote: post #25
taejavu wrote: post #19

Earlier quoted context omitted.

You've misunderstood what you're listening to, I suggest reading the post again. The recordings at the bottom are just recordings of an old lady and a young woman.

Yeah, I understood that. The ones in the middle are generated using their voices. You don't find that amazing?

It's not TEXT to speech, it's speech to speech. I think it would be amazing when we have TTS of that quality.

Re: Deep Voice: Real-Time Neural Text-To-Speech

#42

Post by PieSquared »

slay2k wrote: post #28

How soon before you make an API available? In other words, how do I make use of Deep Voice for my own applications?

Right now, we do not have plans to make an API available. This paper and blog post are mostly meant to describe our techniques to other deep learning researchers and spur innovation in the field. However, we hope that these techniques will be available eventually, and we'll provide more information when that happens.

Re: Deep Voice: Real-Time Neural Text-To-Speech

#43

Post by PieSquared »

modeless wrote: post #8
mrmaximus wrote: post #2

Interesting. They are not TTS like we are accustomed to, they are replicating a specific persons voice with TTS. Listen to the ground-truth recordings at the bottom and then the synthesized versions above. "Fake News" is about to get a lot more compelling when you can make anyone say anything as long as you have some previous recordings of their voice.

> you can make anyone say anything as long as you have some previous recordings of their voice. That's not what this is doing. They're simply resynthesizing exactly what the person said, in the same voice. It's essentially cheating because they can use the real person's inflection. Generating correct inflection is the hardest part of speech synthesis because doing it perfectly requires a complete understanding of the…

Thank you for clarifying this! We tried fairly hard to make this clear, because as you say, the hard part is generating inflection and duration that sounds natural. There's still a ton of work left to do in this duration – we're clearly nowhere near being able to generate human-level speech.

Our work is meant to make working with TTS easier to deep learning researchers by describing a complete and trainable system that can be trained completely from data, and demonstrate that the neural vocoder substitutes can actually be deployed to streaming production servers. Future work (both by us and hopefully other groups) will make further progress for inflection synthesis!

Re: Deep Voice: Real-Time Neural Text-To-Speech

#44

Post by anjanb »

PieSquared wrote: post #16

Hey there! I'm one of the authors of the paper and I'm happy to answer any questions anyone may have! Make sure to check out the paper on arxiv as well.

This sounds cool. What would it take for me to build an Android App with this technology ? Do we have Android/java libraries ?

Re: Deep Voice: Real-Time Neural Text-To-Speech

#46

Post by nojvek »

PieSquared wrote: post #16

Hey there! I'm one of the authors of the paper and I'm happy to answer any questions anyone may have! Make sure to check out the paper on arxiv as well.

How much computing power does this take. When do you see open source implementations running on mobile devices offline?

Re: Deep Voice: Real-Time Neural Text-To-Speech

#47

Post by kayoone »

if i understand this correctly it's a pretty big achievement on the way to being able to replicate any persons voice in the future given enough audio samples. Amazing. Similarly i have seen lip movement (talking) be replicated using machine learning. Having completely artificial (or even real) identities saying whatever you want them to on video is not that far off i guess (simpler than general AI or even fully self driving cars), which is both amazing and terrifying.

Re: Deep Voice: Real-Time Neural Text-To-Speech

#48

Post by ghaff »

dresaj8 wrote: post #4

does anyone know of good ways to do the opposite, speech to text?

Not really. I keep my eye on this area as I generally transcribe my podcasts. But compared to ~$1.50/minute for human transcriptions that require minimal touchup for even fairly tech-heavy topics, nothing I've seen that's purely ML/computer-based comes close to being worth my time to deal with.

Re: Deep Voice: Real-Time Neural Text-To-Speech

#49

Post by PieSquared »

pain_perdu wrote: post #22
PieSquared wrote: post #16

Hey there! I'm one of the authors of the paper and I'm happy to answer any questions anyone may have! Make sure to check out the paper on arxiv as well.

How close (# years?)are we to being able to replicate the voices of any given individual with sufficient samples of their voiceprint?

It's hard to say! We don't quite know exactly how many parameters or minutes of audio are needed to describe fully someone's voice and speaking patterns. Maybe one or two, maybe much more.

Re: Deep Voice: Real-Time Neural Text-To-Speech

#50

Post by PieSquared »

nojvek wrote: post #46
PieSquared wrote: post #16

Hey there! I'm one of the authors of the paper and I'm happy to answer any questions anyone may have! Make sure to check out the paper on arxiv as well.

How much computing power does this take. When do you see open source implementations running on mobile devices offline?

We take several days (2-3) on 8 Titan X GPUs to train our models, which is quite a lot of compute. Running on mobile devices is quite challenging – the inference is not yet fast enough to support that, and has only been optimized for x86 AVX2 CPUs. It may be possible with a fair amount of future work!
Post reply on HN