woodson

5 ms·

31.
by woodson 9mo ago
From your definition: > a learner observes samples from classes which were not observed during training, and needs to predict the class that they belong to. That's not what happens in zero-shot voice cloning, which is why I dismisse
32.
by woodson 9mo ago
I don't disagree, but that's what people started calling it. Zero-shot doesn't make sense anyway, as how would the model know what voice it should sound like (unless it's a celebrity voice or similar included in the trai
33.
by woodson 9mo ago
Does the 169M include the ~90M params for the Mimi codec? Interesting approach using FiLM for speaker conditioning.
34.
by woodson 9mo ago
This generic answer from Wikipedia is not very helpful in this context. Zero-shot voice cloning in TTS usually means that data of the target speaker you want the generated speech to sound like does not need to be included in the training da
35.
by woodson 10mo ago
The local Gemma models are pretty good for tasks involving multilingual inputs (translation, summarization, etc.). They have their niche.
36.
by woodson 10mo ago
There are perceptual hashing algorithms for images/video/audio (dsp and ML based) that could work for that.
37.
by woodson 10mo ago
Not OP, but perhaps they mean not putting too much faith in common benchmarks (thanks to benchmaxxing).
38.
by woodson 10mo ago
> Sometimes DARPA is real slow on identifying good ideas. (more than 8 years later today...) better late than never. Perhaps they noticed a development (breakthrough?) in an adjacent field that makes this now more interesting and/or
39.
by woodson 11mo ago
Looks very similar to Kyutai’s models, given that it uses the same neural audio codec (Mimi) and Depformer module etc.
40.
by woodson 11mo ago
But it’s not like the lower priced models are subsidizing the high-end models (probably the opposite; the high-end ones have greater margins).
41.
by woodson 11mo ago
Just use something like DSPy/Ax and optimize your module for any given LLM (based on sample data and metrics) and you’re mostly good. No need to manually wordsmith prompts.
42.
by woodson 11mo ago
You’re completely right, I misremembered. I edited my post.
43.
by woodson 11mo ago
EDIT: My bad, please disregard; As akreal pointed out, the MMS TTS models aren’t using the SSL models. Original post: You can use the OmniASR SSL models instead of their older MMS models to create TTS models: https://github.com&#
44.
by woodson 11mo ago
The Mac will be much smaller and use less power, though.
45.
by woodson 11mo ago
But that’s the thing, there are no guarantees that the corner cases were actually handled correctly. Especially if it was AI coded without review by a subject matter expert.
46.
by woodson 1y ago
Some countries like Austria have a school type that combines high school with vocational training [1]. It's five years instead of the regular four-year high school there, but you start with 5h/weeks of C programming and write your
47.
by woodson 1y ago
Yes of course. From the outside, I can’t tell to what extent and whether it’s being done sensibly, but there are projects where they are likely better off contracting them out, and others where there are benefits of keeping things in-house
48.
by woodson 1y ago
> So then why not have the same for IT infrastructure instead of outsourcing it to all these parasites? That exists already (and has for a long while): the Bundesrechenzentrum (BRZ, https://www.brz.gv.at/en/ ). They
49.
by woodson 1y ago
> I wish Austria had domestic national IT development teams for national products/websites It (kinda) does: the Bundesrechenzentrum (BRZ, https://www.brz.gv.at/en/ ). They do a lot of public facing government we
50.
by woodson 1y ago
I thought that the effect of these compiler flags was widely known in numerical computing. It allows e.g., reordering of floating point computations and in general disregards IEEE 754. As such, these results are expected, I’d think.
51.
by woodson 1y ago
Nanochat is using PyTorch under the hood. I don’t understand your point.
52.
by woodson 1y ago
Probably NVIDIA’s Parakeet-TDT-0.6b-v3.
53.
by woodson 1y ago
I think the name of the degree (BA/MA etc.) is mostly related to what university or department it is conferred by. For example, if you finish your master studies in Chemistry at the University of Vienna, you get a Master of Arts.
54.
by woodson 1y ago
But that doesn't work for inline edit predictions, right?
55.
by woodson 1y ago
There's an active PR providing inline edit completions via Ollama: https://github.com/zed-industries/zed/pull/33616
56.
by woodson 1y ago
Unfortunately, most big mail providers won’t accept email from your self-hosted mail server, even with DKIM, SPF, etc. So, diversifying is as good as it gets.
57.
by woodson 1y ago
I guess some node.js based tools that are included in Zed (or its language extensions) such as ‘prettier’ don’t behave well in some environments (e.g., they constantly try to write files to /home/$USER even if that’s not your home
58.
Class-action suit claims Otter AI records private work conversations(text.npr.org)
12 points by woodson 1y ago|0 comments
59.
by woodson 1y ago
There are many streaming ASR models based on CTC or RNNT. Look for example at sherpa ( https://github.com/k2-fsa/sherpa-onnx ), which can run streaming ASR, VAD, diarization, and many more.
60.
by woodson 1y ago
Except it’s only been released in September 2022 (not even 3 years ago).

More ›