Adel
@adelmoumen_
Followers
221
Following
2K
Media
7
Statuses
410
AI PhD student @ University of Cambridge & Core Maintainer of @SpeechBrain1. I like realism.
Cambridge, England
Joined February 2020
Delight to announce our new article accepted at #ICASSP2023: Stabilising and accelerating light gated recurrent units for automatic speech recognition. Co-author: @parcollett Code: https://t.co/pplhUFGbx2 Paper:
arxiv.org
The light gated recurrent units (Li-GRU) is well-known for achieving impressive results in automatic speech recognition (ASR) tasks while being lighter and faster to train than a standard gated...
7
5
26
I am a bit surprised I missed the "Neural audio codecs: how to get audio into LLMs" blog post by @vvolhejn. It's really a neat blog post. + a nanoGPT codebase which is always nice [and allows me to compare with my own repo] :) https://t.co/ByE1blOaDj
0
1
2
Today we're putting out an update to the JAX TPU book, this time on GPUs. How do GPUs work, especially compared to TPUs? How are they networked? And how does this affect LLM training? 1/n
38
522
3K
This is *really* good. I am surprised by both the quality and the model’s code-switching abilities. Definitely worth a try!
Kyutai TTS and Unmute are now open source! The text-to-speech is natural, customizable, and fast: it can serve 32 users with a 350ms latency on a single L40S. Try it out and get started on the project page: https://t.co/B4P9FuOrQc
0
0
5
🚀 We're excited to announce our latest work: "Discrete Audio Tokens: More Than a Survey!" It presents a comprehensive survey and benchmark of audio tokenizers across speech, music, and general audio. preprint: https://t.co/QogseMhEGH website:
1
14
35
We are looking for audio and speech generation people, in Zurich, Paris or London to join our team at Google Deepmind. We build cutting-edge speech, music and audio (also audio-visual) generation capabilities. Reach out to Jason or me if interested. Retweets very appreciated !
Our incredible team built many models announced here, including image, voice, music and video generation! And: I'm moving to London this summer, and I'm hiring for research scientist and engineering roles! Our focus is on speech & music in Zurich, Paris & London. DM/email me.
0
9
34
This is the time for hackers to reverse engineer DualPipe algorithms (500 LOC) which imo is the most impressive part of Deepseek technical report. If you don't know how, here is a way to reverse engineer it 👇🔥
🚀 Day 4 of #OpenSourceWeek: Optimized Parallelism Strategies ✅ DualPipe - a bidirectional pipeline parallelism algorithm for computation-communication overlap in V3/R1 training. 🔗 https://t.co/GBtxSvWLT4 ✅ EPLB - an expert-parallel load balancer for V3/R1. 🔗
2
19
159
🗣️🧠 Speech Language Models require lots of compute to train, right? In our new paper, we test is it possible to train an SLM on 1xA5000 gpu in 24 hours? The results may surprise you (they even surprised us)! Tips, open source resources, full paper 👇🏻
7
35
139
And what's cool is you can NOW learn those skills for FREE ! 👇
Distributed ML is probably more useful to learn today than ML. Running 1 model on 1 GPU is a solved problem. The highest-paid people in bigtech are distributed systems experts see https://t.co/n1MVG7Vtwi. Pivot to any software job if ML suddenly collapses.
1
2
50
Today we release Hibiki, real-time speech translation that runs on your phone. Adaptive flow without fancy policy, simple temperature sampling of a multistream audio-text LM. Very proud of @tom_labiausse 's work as an intern.
Meet Hibiki, our simultaneous speech-to-speech translation model, currently supporting 🇫🇷➡️🇬🇧. Hibiki produces spoken and text translations of the input speech in real-time, while preserving the speaker’s voice and optimally adapting its pace based on the semantic content of the
12
53
405
Meet Hibiki, our simultaneous speech-to-speech translation model, currently supporting 🇫🇷➡️🇬🇧. Hibiki produces spoken and text translations of the input speech in real-time, while preserving the speaker’s voice and optimally adapting its pace based on the semantic content of the
21
106
475
🔥 New Speech Recognition Dataset Released! 🔥 The @SpeechBrain1 Team has released 25,000 hours of transcribed and diverse English speech data for both research and commercial use. This dataset is a unified, normalized, and cleaned super set of existing datasets, with tools
3
17
98
Before the release of Picotron data parallel video, a bonus video is out on how to debug distributed codebase ! (Useful as well on a non-distributed settings !)
1
5
16
SpeechBrain version 1.0.2 is now out! My personal contribution is a clean adapters interface that allows custom adapters or integration with PEFT layers, your choice. You can see the tutorial here: https://t.co/Hm50j1obvU
0
2
9
Quantization plays a crucial role in #deeplearning. It speeds up your model and reduces its memory footprint. Curious about how to implement it with #SpeechBrain? Check out Justin Lam's #tutorial: https://t.co/2j5MSGK5qt Explore all tutorials: https://t.co/CkYEluc8Lk
#AI
0
5
15
Funding (€4000) to visit my lab in Toulouse and work with me on combining STT and #pyannote speaker diarization. Please share if you want to see this happen!
10 opportunities available for short lab visits within Europe to help develop new tools to measure early life environments! - part of @COSTprogramme @enviro_dev. Most need some coding experience - but not all. More details here:
3
16
33
Hello Twitter, I am happy to announce that I will be defending my PhD thesis next week! 👨🎓🔔 🕞When? Monday, March 25th, 15h30. 📗Title: Informed self-supervised speech representation learning. 🏫Where? @telecomparis Feel free to DM me for a Zoom link or for details to join!
6
3
56
⚠️ [Appel à Participation] Campagne évaluation DEFT 2024 ⚠️ 📝Tâche : Réponse automatique à des QCM issus d'annales d'examens de pharmacie 🌐 Plus d'infos : https://t.co/bAvXTfe4oF 🚀N'hésitez pas à participer !
0
4
5
Delighted to see the release of #SpeechBrain 1.0 🎉 -- Flower and SpeechBrain have been collaborating since 2021, and together they are the easiest way to federate any speech or audio model. Try out the SpeechBrain➕Flower demo, upgraded especially for today's 1.0 relase by
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: https://t.co/a1wqxLucgw 💻 Repo: https://t.co/MsCZbSbSOf
0
1
7
Excited about the official 1.0 release of @SpeechBrain1!! Super nice to see that our contribution w/ the HyperConformer architecture for compute-friendly low-budget ASR is listed in the ASR section. see more in: https://t.co/rue1PWA1ad abs/paper:
colab.research.google.com
Colaboratory notebook
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: https://t.co/a1wqxLucgw 💻 Repo: https://t.co/MsCZbSbSOf
1
3
16