If I could only give a beginner ML modeler one tip, it would be to prioritize:
• Logging
• Visualizations
• Experiment tracking
…or really anything that makes your results more visible.
That will save you from so many problems.
A critique of ML you sometimes hear from non-ML people is that ML, unlike Physics, lacks certainty, precision, and exactitude. That is by necessity: ML is a fuzzy way to deal with fuzzy problems.
"Let's go to the zoo" 💬
Check out this new model repository for ASR in many languages! That is something we don't see enough in the ASR community... well done @coqui_ai !! 👏🚀
🔥🔥 Coqui community model zoo is Open 🔥🔥
✅ Speech-to-Text models for 36 languages
✅ Models for on-device (@TensorFlow Lite)
✅ Easily customize with your own Language Model
Kudos to the model creators and community members who made this possible 🎉
https://t.co/D8i3l8yfnb
This #neurips2020 workshop should be really fun: "I Can't Believe It's Not Better! Workshop": beautiful and elegant ideas that turn out not to work in practice https://t.co/VDs6Gmyr42
Happy to have had the chance to present a keyword spotting project I loved working on and to have been part of such an interesting technical program @interspeech20. Here is what we learned at the conference: https://t.co/dfTBRSIrF6
Attending scientific conferences has become a whole new experience. We have to rethink how to make the most of the virtual format, which opens a world of new exciting possibilities. We share our experience as attendees of @interspeech20 here:
https://t.co/x2kpLjzjXg
In this thread, I’m sharing some recent work published in 2020 by the @Sonos Voice Experience team.
While I can't wait for 2021, these are among the few good things that happened this year 🧵👇
Ne devrait on pas avoir accès aux arguments détaillés de ces scientifiques qui conseillent le gouvernement et préconisent entre autres le maintien des élections ?? 🤔 d’habitude c’est comme ça qu’on fait en sciences : on expose ses théories.
We are pleased to announce CoVoST, a multilingual speech translation corpus based on Common Voice. It has English translations for 700h speeches in 11 languages (with 11K+ speakers). #LREC2020@juanmiguelpino@thoma_gu@facebookai
https://t.co/WwCyTE619O
https://t.co/zsrbHjbh87
#Distinction🥇
Laurence Devillers, Professeure en Informatique à Sorbonne Université et chercheuse au @Limsilab, experte en Intelligence Artificielle et en éthique de la recherche en robotique a été nommée Chevalier de la Légion d'honneur. Félicitations !
New year, new adventures?
Join the @sonos Voice team (ex-@snips) to build a better and more private Voice Experience for millions of users !
We have 5 positions open, a great team, and more ambitions than ever…
https://t.co/2eH1pl0G16
Contact me for more details!
Happy to announce the latest release from @facebookai for self-supervisied learning: Libri-light, a benchmark for the training of automatic speech recognition systems with limited or no supervision. https://t.co/k7UIKFZEmA