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

21–30 of 80 posts

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

#21

Post by mbrookes »

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.

> "Fake News" is about to get a lot more compelling hen you can make anyone say anything as long as you have some previous recordings of their voice.

Adobe has already developed that technology:

https://arstechnica.co.uk/information-technology/2016/11/ado...

Now imagine combining it with this:

Face2Face: Real-time Face Capture and Reenactment of RGB Videos https://www.youtube.com/watch?v=ohmajJTcpNk

Perhaps using the intonation from the face-actor's voice to guide the speech synthesis.

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

#22

Post by pain_perdu »

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?

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

#24

Post by qq66 »

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?

"My voice is my passport."

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

#25

Post by monk_e_boy »

taejavu wrote: post #19
monk_e_boy wrote: post #3

OK, that went from uncanny valley to flipping amazing. I could picture the person speaking. An old lady. A young woman. It was hard to picture an algorithm in a machine. It's amazing that is all boils down to 1s and 0s and some boolean logic.

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?

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

#26

Post by skykooler »

dresaj8 wrote: post #14
forthefuture wrote: post #6

Earlier quoted context omitted.

Depends on how good you're talking. Chrome supports the SpeechRecognition API. https://developer.mozilla.org/en-US/docs/Web/API/SpeechRecog...

i'm more thinking of ways to programmatically turn long audio files into indexable text.

Julius[1] can do this. But the accuracy depends on the language model you are using, and unfortunately the free English language model (VoxForge) is not the best.

[1] http://julius.osdn.jp/en_index.php

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

#27

Post by Smerity »

methou wrote: post #23
hprotagonist wrote: post #7

how does this stack up against wavenet?

It's in the abstract. "... For the audio synthesis model, we implement a variant of WaveNet that requires fewer parameters and trains faster than the original ..."[1] [1]: https://arxiv.org/abs/1702.07825

Disclosure: I'm one of the co-authors of the QRNN paper (James Bradbury, Stephen Merity, Caiming Xiong, Richard Socher) produced by Salesforce Research.

There are many interesting advances that Deep Voice paper and implementation make but the part I'm excited by (and which might be transferable to other tasks that use RNNs) is showing that QRNNs are indeed generalizable to speech too - in this case in place of WaveNet.

"WaveNet uses transposed convolutions for upsampling and conditioning. We find that our models perform better, train faster, and require fewer parameters if we instead first encode the inputs with a stack of bidirectional quasi-RNN (QRNN) layers (Bradbury et al., 2016) and then perform upsampling by repetition to the desired frequency."

QRNNs are a variant of recurrent neural networks. They're up to 16 times faster than even Nvidia's highly optimized cuDNN LSTM implementation and give comparable or better accuracy in many tasks. This is the first time that it has been tried in speech - to see them note the advantages hold across the board (better, faster, smaller) is brilliant!

If you're interested in technical details, our blog post[1] provides a broader overview and our paper is available for deeper detail[2].

[1]: https://metamind.io/research/new-neural-network-building-blo...

[2]: https://arxiv.org/abs/1611.01576

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

#29

Post by bmc7505 »

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.

Hi Andrew, congratulations on your result! A few questions, feel free to answer one or any. How close do you think you are to having fully end-to-end models for speech? Are you optimistic we can get speech synthesis to run on mobile devices in the near future? Do the inference optimizations (particularly sample embedding and layer inference) generalize well to other architectures, like speech recognition? It seems that if these models are going to run offline in realtime on mobile devices, we will need to have specialized hardware, but maybe we can squeeze enough performance out of mobile CPUs to get a highly optimized version to work. Thanks!

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

#30

Post by gcr »

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?

I mean, it's sort of amazing, but it wasn't completely generated by machine. Those sound clips in the middle were generated by copying the inflections from actual recordings, not generating the inflections from scratch. It sounds like the current system they have sounds like the robotic voices at the very top.
Post reply on HN