Baidu Research Blog

I think they still have offices here in the US and employ Americans. Alibaba has been scaling back though. Lastly, I don't think there's any law that prevents an American from moving to China to work. https://careers.tencent.com/en-us/cityinfo.html?city=401 http://research.baidu.com/

On the contrary, Graph NN's I've found Graph NNs extraordinarily useful and very easy to train! I'm not sure what kind of problems you are trying to use them for, but generally they are really useful for node classification or recommendation type tasks. So for example in the knowledge graph context, you can do things like give it a tiger, jaguar and panther and it will find nodes like lions and leopards. https://arxiv.org/pdf/1901.00596.pdf is a survey paper which has a decent overview of the tasks they are useful for. http://research.baidu.com/Public/uploads/5c1c9a58317b3.pdf is using them for QA, but I'm not familiar enough with the dataset to evaluate it fairly.

Thank you so much for this link, that is the best text-to-speech with an open architecture I've ever heard 'til now. Under https://github.com/keithito/tacotron you can find a pre-trained model based on this paper, although it isn't matching the quality yet. Maybe I can get some cluster time to train a new model using multiple datasets. Edit: Another interesting one: http://research.baidu.com/deep-voice-3-2000-speaker-neural-text-speech…

Sounds amazing! Just to add a usecase - for many people, creating a decent voiceover is one of the big sticking points for producing youtube videos or educational courses. If I could write a script, and have software generate a decent enough voiceover, it would be amazing. It's not even necessary to copy anyone's voice, as long as there's a selection of the most comprehensible and human-sounding ones. Then, you could even automatically generate slideshow presentation from a few illustrations and headlines, and that would make "rendering" articles into videos very fast and easy. I'm sure a lot of people would pay for such service. ---- By the way, recently I've encountered Deep Voice 2, a similar research project by baidu: http://research.baidu.com/deep-voice-2-multi-speaker-neural-text-speec… Results are very impressive.

This post on Deep Voice seems a little off-the-mark. In fact, I would say it is completely misleading about the technical accomplishments here. From my perspective, Baidu's approach is a little embarrassing, with the use of many modeling stages in their training and production of TTS. When the rest of the community is moving towards end-to-end training, their usage of this many stages sounds excruciating. Merlin[0], which was a pretty good standard for 2016, has this painful feeling as well, with two DL stages (duration, acoustic) followed by some conditioning and then a synthesis step. The more important technical contribution seems to be the hand-tuned synthesis code that makes their generation faster; cool but not particularly sexy (and there are few details). The details on training hyperparams are nice to have too, of course. Contrary to the post, I would be very surprised if the voice sample included in the post was actually generated by Deep Voice -- it has none of the robotic qualities pointed out by the researchers themselves in their blog post[1]. More likely it is a demonstration of the loss in their last, WaveNet-like stage. This was also pointed out in the previous HN discussion[2] Lastly, Andrew Ng is neither thanked in the paper nor mentioned on any webpage -- are we sure this was work he supervised? [0] https://github.com/CSTR-Edinburgh/merlin [1]…

What's interesting about Baidu's AI model Ernie is that Baidu and its founder, Robin Li, have been working on AI for a long time. Robin Li has a strong background in AI research going back many years. Also notable is that some of the key early research on scaling laws—important for understanding how AI models improve as they get bigger—was done by Baidu's AI lab. This shows Baidu's significant role in the ongoing development of AI. https://research.baidu.com/Blog/index-view?id=89 I am excited to see Baidu catchup. It feels like they have earned it. Being very early.

Also left out the Baidu scaling laws paper from 2017, and his circle has a history of a kind of citation ring type thing leaving earlier stuff out https://research.baidu.com/Blog/index-view?id=89

Their backend originates from Baidu ERNIE: http://research.baidu.com/Blog/index-view?id=160

Yes, certainly. One industry use-case that comes to mind if Baidu; white-paper link below [1]. [1] http://research.baidu.com/Public/uploads/5e76df66c467b.pdf

DeepSpeech 3 from Baidu also uses transducers. http://research.baidu.com/Blog/index-view?id=90.

Can you cite some sources on the voice recognition part? I seem to remember that Google(https://research.googleblog.com/2015/08/the-neural-networks-behind-goo… and Baidu(http://research.baidu.com/deep-speech-3%EF%BC%9Aexploring-neural-trans… both use neural networks for speech transcription. I don't really believe that the implementations out of the startups are meaningfully/significantly better than the implementations out of Google/Facebook. If they were seriously better they'd be acquired (as we've seen over and over again).

"Deep Voice 2 can learn from hundreds of voices and imitate them perfectly. Unlike traditional systems, which need dozens of hours of audio from a single speaker, Deep Voice 2 can learn from hundreds of unique voices from less than half an hour of data per speaker, while achieving high audio quality." - http://research.baidu.com/deep-voice-2-multi-speaker-neural-text-speec… Now imagine that with Tacotron quality, and you'll get that "strange" effect with anyone, meeting their vocal clone. This is still text-to-speech, so it's not live-copying your intonation, but you could easily imagine a seq2seq network designed to do so.

Well, the question is, do they just need to throw more computational power / training at this algorithm or is that the peak of their implementation? This is something Google has been working a lot on [1] and Baidu also recently posted about their results too [2]. We're definitely pretty close to passing the human detectable level. [1] https://deepmind.com/blog/wavenet-generative-model-raw-audio/ [2] http://research.baidu.com/deep-voice-production-quality-text-speech-sy…

The more interesting would be to see how feasible is to crack the in band SAS authentication string, when callers verbally verify it. Deep learning and ability to train on a specific callers' voice [1] then mimic it might be an interesting attack vector. In practice Silent Circle's implementation does something interesting and instead of SAS numbers use dictionary words. So you end up with something like "Pink Elephant Salad". Could probably MitM that. However callers are then supposed to make some extra puns or discuss it a bit and say something like "Ha-ha! Wonder how tasty the an elephant salad would be". And if after MitM-ing, the string to the other side was "Plastic Blue Llamas" then a MitM attack becomes more obvious. [1] http://research.baidu.com/deep-voice-production-quality-text-speech-sy…

We had a discussion about this http://research.baidu.com/deep-voice-production-quality-text-speech-sy… six days ago, with demos. This post links the arxiv paper instead. Old discussion at https://news.ycombinator.com/item?id=13756489

Hey there! Our paper is on Arxiv [0] and we have a blog post about this [1]. I am one of the authors on this paper and am happy to answer any questions you may have. [0] https://arxiv.org/abs/1702.07825 [1] http://research.baidu.com/deep-voice-production-quality-text-speech-sy…

The samples are here - http://research.baidu.com/deep-voice-production-quality-text-speech-sy… For reference, other recent TTS implementation samples: WaveNet - https://deepmind.com/blog/wavenet-generative-model-raw-audio/ SampleRNN - https://soundcloud.com/samplernn/sets And Char2Wav - http://www.josesotelo.com/speechsynthesis/

Audio samples here. http://research.baidu.com/deep-voice-production-quality-text-speech-sy…

Blog post: http://research.baidu.com/Blog/index-view?id=106

It's unfortunate that there aren't any high quality projects for the other way around - text to speech. By high quality I mean something as good as Tacotron 2 [1]. However, I have no illusions of Google ever releasing code or trained models. Someone will likely re-implement it from the paper, but it's unlikely to be anywhere near as good as those samples, due to worse tuning and lack of a good training set[2]. This effect can be seen if you compare Baidu's samples [3] for Deep Voice 3 with those from an implementation by a third party [4]. [1]: https://google.github.io/tacotron/publications/tacotron2/index.html [2]: Common voice, for example, will not work well, since you want lots of data from a single speaker for good results. See for example, how samples get progressively worse when increasing the number of speakers in training data[3]. [3]: http://research.baidu.com/deep-voice-3-2000-speaker-neural-text-speech… [4]: https://r9y9.github.io/deepvoice3_pytorch/