The #SpeechBrain team is working hard to release soon an open-source all-in-one toolkit based on #PyTorch specifically designed for multiple tasks (e.g., #ASR, enhancement, separation, speaker recognition/diarization, multi-mic processing). speechbrain.github.io milamontreal

#SpeechBrain is growing! Today, let me share a tutorial on "Speech Classification from Scratch". Helpful for speaker-id, language-id, emotion recognition, sound classification, etc... Tutorial: colab.research.google.com/drive/1UwisnAj… Web: speechbrain.github.io Code: github.com/speechbrain

I've used it for my own research. Feel good 😇.

#SpeechBrain is growing very fast! We are happy to announce the new call for #sponsors. We have very ambitious plans for the future and sponsors can play a crucial role in the development of our #Opensource #toolkit. More info here: speechbrain.github.io/img/Call_for_S…

Researchers Had To Shut Down AI After It Taught Itself 19 Languages?!* 🤔😱🤖😤 Like👍 Subscribe🔔 * = we used pseudo-labeling to train a single massively multilingual speech recognizer for all 60 languages of Common Voice. Paper: arxiv.org/abs/2111.00161 🧵

Can we summarize text to support a decision? How does that differ from text-only summarization? We proposed a novel task of decision-focused summarization which aims to generate summaries for a target decision. #EMNLP paper: arxiv.org/pdf/2109.06896… video: youtube.com/watch?v=by0JJH…

Receive the YouTube Creator Silver Award for 100,000 subscribers. When I started uploading videos about DL to YouTube in the fall of 2016, I never imagined this achievement. Thanks to all subscribers. We learn together. youtube.com/c/HungyiLeeNTU/

▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 101%

I find it quite surprising how much smaller diffusion models tend to be versus transformers generative model of similar quality. Where could this come from? Better parameter sharing, different scaling law behavior, better use of the source of randomness, smt else…?

(1/5)🚨Can LLMs be more factual without retrieval or finetuning?🤔 -yes✅ 🦙We find factual knowledge often lies in higher layers of LLaMA 💪Contrast high/low layers can amplify factuality & boost TruthfulQA by 12-17% 📝arxiv.org/abs/2309.03883 🧑‍💻github.com/voidism/DoLa #NLProc

Introduce In-Context Pretraining🖇️: train LMs on contexts of related documents. Improving 7B LM by simply reordering pretrain docs 📈In-context learning +8% 📈Faithful +16% 📈Reading comprehension +15% 📈Retrieval augmentation +9% 📈Long-context reason +5% arxiv.org/abs/2301.12652

Please let's rethink quality evaluation of generative models, especially for audio!

Exciting news! 🎉 #SpeechBrain 1.0 is out with tons of thrilling advancements. Our #OpenSource toolkit now features 200+ recipes and 100+ pretrained models on #HuggingFace for diverse #ConversationalAI tasks. 🌐 Website: speechbrain.github.io 💻 Repo: github.com/speechbrain/sp…

Nathan MrBusiness Scientific knowledge is rarely secret. Technical knowledge doesn't remain secret for very long. Know-how and practical expertise takes longer to disseminate. Product technology and market experience takes even more time and money to reproduce.

I rarely agree with Yann on his take on LLMs. But I strongly agree with him on this one.