Viewing profile — PieSquared

Probably worth adjusting the title -- air quality is not just a measure of pollution. In this case, both SF and Portland are filled (unfortunately) with wildfire smoke, rather than…

Are you folks planning on extending this to speech? I'm always been disappointed by how speech vocoder networks aren't built with any great inductive biases for waveform generation…

I don't think anything about the current set of tools is specific to sample rate; WaveNet, Tacotron, WaveRNN, etc, should work fine to generate 44.1Khz audio. They might just need …

There's a few recent papers actually that show minor improvements by integrating LPC prediction into deep methods ([0], [1]). In my experience (some of which comes from reproducing…

First of all, it's important to note that Tacotron and WaveNet are responsible for different parts of the speech synthesis pipeline, so the comparison here isn't quite accurate. Sp…

I'll try to answer these one at a time. 1. Does text-to-speech require AI? This one is a bit tricky to answer since it requires defining "AI". AI as a moniker has been used to desc…

I'm an author on a few of these papers referenced (the Deep Voice papers from Baidu). I'm happy to answer any questions folks may have about neural speech synthesis, as I've been w…

Ta-Nehisi Coates, Between the World and Me. https://en.wikipedia.org/wiki/Between_the_World_and_Me "It is written as a letter to the author's teenage son about the feelings, symbol…

The new Gmail interface has snooze, so if that's the only thing you're missing it'll stick around. I mostly miss the "Trips" feature, having recently switched from Inbox to new Gma…

"Monoid" is an adjective that describes a data type. Anything you describe as a monoid has to have three properties: you can add them together, there's an "empty" or "zero" value, …

In addition to the blog post, there's an interesting discussion on Github that happened before the blog post was published: https://github.com/Unikernel-Systems/unikernel.org/pull/…

To be fair, this article is really pretty "cutting edge" as far as Haskell goes. It describing a new feature they're planning on adding to the Haskell compiler, one that significan…

Feel free to get in touch for more Q/A, my email is in my profile. We've experimented a bunch with many of these hyperparameters. Our phoneme signal has mostly stayed 256 Hz, but w…

I don't quite know what VoCo does, but it seems like a concatenative system that they've tuned a huge amount. I'm a little skeptical that it works as well and as reliably in real l…

Check out Char2Wav (recent) and SampleRNN (the RNN-based audio synthesis architecture). The related work section of the Deep Voice paper mention a bunch of related papers that are …

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 en…

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 mu…

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…

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 innovatio…

Baidu as a company doesn't use TensorFlow (as far as I know). We have our own high-performance and easy to use open source framework called PaddlePaddle [0], which is quite powerfu…

Thank you! For fully end-to-end models, it's hard to say exactly. The Char2Wav paper demonstrates that there is hypothetically an architecture and a set of weights that can do synt…

We didn't actually try LSTMs, because we train in 1.25 second chunks, so running an LSTM for several hundred timesteps would drastically slow down training. Our per iteration time …