kastnerkyle

7 ms·

1.
by kastnerkyle 2y ago
The direct counter-argument to "worst representation" is usually "representation with fewest assumptions", waveform as shown here is getting close. Though recording environment, equipment, how the sound actually gets dig
2.
by kastnerkyle 2y ago
On the loops / sampling front: I always thought RAVE [0][1][2] was a very interesting approach, that really embraces latent spaces and sample/stretch type approaches in the waveform space [0] https://github.com/aci
3.
by kastnerkyle 2y ago
Research into "pure" unconditional generation can often lead to gains in the conditional setting. See literally any GAN research, VQ-VAE, VAE, diffusion, etc - all started from "unconditional/low information" pretty
4.
by kastnerkyle 3y ago
This work is another classic in the "neural nets meet spreadsheets" genre [0]. Really helps visualize what is going on in (at least some) latent spaces. [0] https://vusd.github.io/spacesheet/
5.
by kastnerkyle 3y ago
This same technique, extended can work well for detecting plagiarism from the underlying corpus as well, by tracking a trie of "good" completions in the n-gram sense, and a longer trie of "no-good" completions. This tech
6.
by kastnerkyle 3y ago
There was a really nice post on doing this kind of thing with CRF back in 2015 [0]. Open source data, and code on github. Also a nice tutorial on structured prediction using CRF type models. Would be interesting if you could prompt, LoRA di
7.
by kastnerkyle 3y ago
What since Adam? Learning rate scales / schedules? I cannot think of many big massive changes since ~2014, most of the setups from that era (grad clip + medium-ish LR, some ramp up or roll-off at the end) work fine today for me. (Note:
8.
by kastnerkyle 3y ago
I disagree with this. Binarized MNIST samples of any reasonable quality are (still) tricky to get right without a hierarchical system (read: VQ-VAE tokens or some such encoder space). Same with really solid CIFAR-10. "Scaling down&quot
9.
by kastnerkyle 3y ago
Previously TortoiseTTS was associated with PlayHT in some way, although the exact connection is a bit vague [0]. From the descriptions here it sounds a lot like AudioLM / SPEAR TTS / some of Meta's recent multilingual TTS app
10.
by kastnerkyle 3y ago
Maybe 'Image Quilting for Texture Synthesis and Transfer', Efros and Freeman [0]? There's some neural / patch blends from 2016 that I always thought were interesting (CNN-MRF) [1], and I think there's a renaissance
11.
by kastnerkyle 3y ago
For any models where you get access to the code (counting HF in this), definitely a lot of tooling available - and this tool looks really nice hadn't seen it before! Hope similar levels of control can become available for the bigger, l
12.
by kastnerkyle 3y ago
Discrete diffusion with rewriting can work well. It feels loosely similar to backtracking, if you assume n_steps large enough - need to be able to rewrite any non-provided position though I think (not all setups do this). Downside is the
13.
by kastnerkyle 4y ago
You might be interested in this thread from 2019 here on HN as well, I linked to a line of work on constrained sequence generation from Sony CSL that is pretty nifty https://news.ycombinator.com/item?id=21334639 As an aside
14.
by kastnerkyle 4y ago
On this point, for images much of the structure is encoded in the phase . This fact was very surprising to me when I learned it. See the images in this stack overflow for an example [0]. [0] https://stackoverflow.com/a&#x2F
15.
by kastnerkyle 4y ago
I really like the descriptions from SUNDAE ( https://arxiv.org/abs/2112.06749 ) if you have some background about general neural net style modeling, and generally find the multinomial or binomial diffusion settings a bit
16.
by kastnerkyle 4y ago
I've been trying to follow your project since I heard about it - but hadn't read the link you posted here. Really thorough, really good description of many of the quagmires (with prior cases) of melody and music copyright. Thanks!
17.
by kastnerkyle 4y ago
https://www.theatlantic.com/technology/archive/2020/02/whats... is an interesting related article on this topic
18.
by kastnerkyle 4y ago
Also, if you are seeking more melodies / catchy tunes to model NES-MDB ( https://github.com/chrisdonahue/nesmdb ) might fit the bill. The modeling paper using that dataset + transformer XL - LakhNES ( https:/&#
19.
by kastnerkyle 4y ago
I've had pretty good luck recently with a mix of SUNDAE ( https://arxiv.org/abs/2112.06749 ) and coconet ( https://arxiv.org/abs/1903.07227 ) and/or Music Transformer based internal models r
20.
by kastnerkyle 7y ago
Yes - the crux is just to add some logic and throw out beams which don't match your constraint, then rank candidates based on sequence probability. You can roll-back the generation process and/or mask the probability distributio
21.
by kastnerkyle 7y ago
There are many flavors of beam search - I have found that for adding explicit checks and constraints (for example rhyme constraints or certain pivot words) the resulting proposals are generally a lot better. Even with simple markov chains I
22.
by kastnerkyle 7y ago
Tricks in beam search to force rhyme schemes, or techniques like constrained markov chains (c.f. https://redylan.neocities.org/#/how-it-works/ and https://github.com/gabrielebarbieri/markovcha
23.
by kastnerkyle 8y ago
Much less audio is potentially needed for TTS than ASR, however the spread and quality of the TTS dataset is critical which is one reason why just training on ASR datasets "in reverse" hasn't worked great. For example, commer
24.
by kastnerkyle 8y ago
Worth noting that a big chunk of the core TTS code here is built on tools from other researchers like Ryuichi Yamamoto and Keith Ito, and they have great implementations to check out as well. The best quality I have heard in OSS is probabl
25.
by kastnerkyle 8y ago
This type of concatenation I first saw in Alex Graves' work on "Generating Sequences With Recurrent Neural Networks", including his unpublished TTS demo [1]. Biasing with part of another sentence (as in handwriting) can possi
26.
by kastnerkyle 8y ago
Glad to see people posting links to music modeling projects - it's a huge space, and a ton of fun to actively work on or just mess with if you like music and code. Even simple markov models can get someplace with a bit of tweaking [0][
27.
by kastnerkyle 8y ago
See [0] for a method combining branch and prune with Markovian probabilities. I did a hacky version of my interpretation of this work at [1]. [0] https://link.springer.com/article/10.1007/s10601-010-9101-4 [1] ht
28.
by kastnerkyle 8y ago
I don't think this is necessarily true. CNNs can produce incredible results, see for example [0]. A snippet from the paper (their code is also available): """ Prompt: The Mage, the Warrior, and the Priest Story: A light
29.
by kastnerkyle 8y ago
I've had good luck with Caverphone for a number of speech specific tasks [0]. There is a python implementation directly in the pdf, I also wrote a version here [1], no idea if it exactly matches the pdf version but it worked for my cas
30.
by kastnerkyle 8y ago
It may be possible to do this with an LSTD VAD, I always had really good luck with that. I tried a few random ones in here for silence removal - no quality guarantee [0] I found LTSD pretty robust compared to simpler energy based things as

More ›