woodson

16 ms·

241.
by woodson 5y ago
Local taxes are usually not an issue if hired as a contractor. You will have to pay taxes and social insurance system contributions in the local country according to the local laws yourself. Getting hired as a regular employee is definitely
242.
by woodson 5y ago
If you're fine writing code to build a solution yourself, try the Conformer or ContextNet models in Nvidia NeMo ( https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en... ) or Vosk ( https:&#x2
243.
by woodson 5y ago
Well, in some disciplines PhD theses can be basically a few journal or conference papers stapled together (with intro and conclusion tying things together). At least there would likely been more scrutiny during the publication process of th
244.
by woodson 5y ago
NVidia NeMo: https://github.com/NVIDIA/NeMo
245.
by woodson 5y ago
Vosk is built on Kaldi.
246.
by woodson 5y ago
It's somewhat more difficult for founding a limited liability company there, where the capital requirements are much higher than in most other countries (Eur 35K).
247.
by woodson 5y ago
Yes, but then you don't need Mozilla collecting read speech samples. You can just scrape any audio out there, run speech activity detection, and there you go.
248.
by woodson 5y ago
Then you need a lot of people that listen to those 12B hours of audio, and multiple listeners agree for each chunk of audio that what is spoken corresponds to the transcript.
249.
by woodson 6y ago
All non-touchscreen XPS models I've seen were matte.
250.
by woodson 6y ago
Eduworks | Machine Learning Engineer (Speech) | Fully remote | Full-time | US Citizen or Green Card We are an R&D company located in Corvallis, Oregon. We are developing a low-latency speech / spoken language understanding based pr
251.
by woodson 6y ago
But international PEOs have such crazy high fees that finding a local tax accountant suddenly becomes much less of a bad idea.
252.
by woodson 6y ago
In general, no, especially for typical data analysis tasks. MATLAB used to have an edge in signal processing, but (IMHO) lost it to Python-based solutions (especially when running things on GPUs, ML or not). If you're doing a lot of ex
253.
by woodson 6y ago
Backup your profile before the update. Thunderbird 78 makes non-backward-compatible changes to it, so you cannot downgrade to TB68.
254.
by woodson 6y ago
This is somewhat off-topic, but as someone who has worked on various speech processing and ASR projects, I'm curious to learn from people who have specific problems and applications for this technology where it can make a difference. T
255.
by woodson 6y ago
You can also check out https://github.com/TensorSpeech/TensorFlowASR for inspiration (not my project, not involved). It implements streaming transformers and conformer RNN-T (but in TF2). Deployment on device as TFLite
256.
by woodson 6y ago
Good for you, it's the only way to know which tool works best in your case. I did the same for my use case and arrived at the opposite conclusion. What most people don't realize is that it heavily depends on your use case and doma
257.
by woodson 6y ago
Check out Kaldi. It's a toolkit rather than a ready-to-deploy service but has some solid pretrained models and recipes for training your own. You can use various existing projects for deployment, e.g. vosk-server (also for on-device) w
258.
by woodson 6y ago
Unfortunately, nvidia-settings requires xorg to be installed and running (even when run in cli mode), and it's pretty much the only way to control fan speeds on consumer GPUs.
259.
by woodson 6y ago
I believe it also supports what it calls "pulse" models converted/imported from Kaldi for low-latency ASR/keyword spotting (though this may have changed, I haven't checked in a while).
260.
by woodson 6y ago
Fully agreed, taxes and work permit/visa are topics that are likely being neglected by most. Often, there simply isn’t a visa category that allows individuals to work for foreign entities while residing in another country, as regular w
261.
by woodson 6y ago
Staying at AirBnBs in the listed cities and working from there is quite likely cheaper than renting office space in Boston (probably even including airfare). So it’s not about “resources” as in money, it’s about being able to be productive
262.
by woodson 6y ago
https://parmarsuraj99.github.io/suraj-parmar/jupyter/nlp/hug...
263.
by woodson 6y ago
In a certain way, they could be, simply because their structure may work better on the type of language that English is. It may not work as well for languages that exhibit other grammatical patterns (e.g. morphologically rich languages, or
264.
by woodson 6y ago
Any chance you would be willing to share the (finetuned) models? For science? I’m interested in detection and want to look at artifacts common to speech audio generated by different models all based on the same overall implementation.
265.
by woodson 6y ago
I like to see the other side (that of the true believer) as well. If all (or most) can agree on that it is the right thing to do and the current way of doing it is wrong, why oppose the change for 20 years saying it will cause breakage? If
266.
by woodson 6y ago
Perhaps they’ve been using the WaveSurfer software for audio analysis ( https://en.m.wikipedia.org/wiki/WaveSurfer ).
267.
by woodson 7y ago
ESPnet apparently already has implementations of versions of transformer transducers with RNN-T loss (though, with different network architecture). At least the paper cites results on a freely available dataset, right?
268.
by woodson 7y ago
Agreed, and that's where it seems having lots of experience working with speech data helps more than trying to brute-force it with just larger CTC models and "more" data of dubious quality.
269.
by woodson 7y ago
Recent work on transformer transducers with limited right (and left) context seem to give decent results as well: https://arxiv.org/abs/2002.02562
270.
by woodson 7y ago
Hi Alexander, I appreciate you posting here; I didn't feel too strongly about this to sign up for an account to comment directly on TheGradient. > Kaldi cool, NNs not cool It seems you're misrepresenting things. Kaldi models a

More ›