Sung-Lin Yeh

PhD student @ University of Edinburgh. Previously: @AIatMeta

[Sung-Lin Yeh @30SecToMercury]

@unilightwf @hidekikawahara @erica_cooper Also found this review useful.

[Sung-Lin Yeh @30SecToMercury]

For Q4 (Why dMel?), we did observe a gap between optimizing the encoder w/ mel-filterbanks v.s. dMel in our recent paper:
https://t.co/J31quG9yKX

Not sure the FSQ in their dMel is actually optimized w.r.t. the training objectives tho.

[Desh Raj @rdesh26]

https://t.co/cKfx1zYfXY

[arXiv Sound @ArxivSound]

Sung-Lin Yeh, Wei Zhou, Gil Keren, Duc Le, Zhong Meng, Hao Tang, Jay Mahadeokar, Ozlem Kalinli, Alexandre Mourachko, "MELD: Mel-Spectrogram-Based Speech Language Modeling with Discrete Latent Variables," https://t.co/oFFz6LRQD6

[Sung-Lin Yeh @30SecToMercury]

@unilightwf It's done on-the-fly in speechbrain:
see here:
https://t.co/ZkU3RVOw4T
and the config:
https://t.co/WSQp3ssIMa

[Dileep George @dileeplearning]

Check out the recent updates to PGMax, our JAX package for implementing large discrete probabilistic graphical models. It can now solve the smooth dual of LP relaxation of MAP problems orders faster than LP solvers, & more 1/6
https://t.co/sx9Zwyrb6t

[Sung-Lin Yeh @30SecToMercury]

My paper on learning discrete speech representations with neural HMMs was accepted at ICASSP23 🎉

paper:
https://t.co/hhTegkp11U
The hmm layer:
https://t.co/k9BnwC4aD4

[YCY @yoyolicoris]

Happy to announce this work, an unsupervised speech super-resolution algorithm based on posterior sampling in diffusion models, was accepted at @ieeeICASSP !
We thank all reviewers for reviewing and giving high recognition.

[Sung-Lin Yeh @30SecToMercury]

@Li_Yuan_Chao give up

[Sung-Lin Yeh @30SecToMercury]

New preprint with @yoyololicon!

[YCY @yoyolicoris]

A new way to do audio super-resolution with diffusion models!
Teamed with @30SecToMercury, we introduced a diffusion sampling algorithm for speech super-resolution.
https://t.co/DShVJWcsVY

Demo page and source code (with pre-trained models!): https://t.co/UCZVkqrwa8

[Sasha Rush @srush_nlp]

Autodiff Puzzles (https://t.co/pAQh3BcGd8)

20 spooky short puzzles that will help you understand tensor autodiff.

[François Fleuret @francoisfleuret]

A recent exchange on Twitter made me realize that many may be using tensors without knowing how they are represented in memory, and how it makes some operations super fast.

This is standard stuff common to many tensor libraries, including our beloved @numpy_team and @PyTorch.

[Sung-Lin Yeh @30SecToMercury]

My first PhD paper was accepted at INTERSPEECH 2022, an unsupervised co-training model with discrete latents that unifies various self-supervised training methods (e.g., HuBERT).

Autoregressive Co-Training for Learning Discrete Speech Representations
https://t.co/v7g2KxrkQy

[Andrew Davison @AjdDavison]

Can anyone explain if there is a difference between unsupervised and self-supervised learning? To me they seem the same and I find myself using both terms interchangeably (I prefer unsupervised), but I feel like I'm confusing people who understand them to mean different things.

[Sung-Lin Yeh @30SecToMercury]

Glad to be part of it!

[Mirco Ravanelli @mirco_ravanelli]

I'm happy to announce that a preprint paper on #SpeechBrain is now available on #arXiv:

Preprint: https://t.co/qibipubtfW
Website: https://t.co/a1wqxLucgw

That's a good read for the weekend after #ICASSP2021! 😄

@PyTorch @huggingface #DeepLearning #MachineLearning #AI #Speech

[Leo Boytsov @srchvrs]

Huge news,
@Microsoft
buys the speech recognition company Nuance, which is the biggest competitor of my former employer. Nuance is purchased for nearly $20B, which seems to be a 2d biggest after
@LinkedIn
https://t.co/gNxjRmvSK0

[Mirco Ravanelli @mirco_ravanelli]

I'm thrilled to announce the public release of #SpeechBrain!

#SpeechBrain is an #opensource toolkit designed to make research and development of #speech technologies faster. It is flexible, modular, easy-to-use, well documented.

https://t.co/a1wqxLucgw

#DeepLearning #AI

[William Falcon ⚡️ @williamfalcon]

Excited to launch torch-metrics - @PyTorch metrics optimized for distributed training at scale.

We thought the lightning metrics could be useful to all @PyTorch users beyond just @PyTorchLightnin :)

https://t.co/CbFVEb2VGy

[Yann LeCun @ylecun]

A new blog post I wrote with Ishan Misra.
An overview of Self-Supervised Learning.

We look at recent progress in SSL for vision & explain why SSL is more challenging with high-D continuous signals (images, video) than it is for discrete signals (text).

https://t.co/DlL885CPpb

[Kyunghyun Cho @kchonyc]

students are curious, and their questions are great: a few select questions from this semester, and my (incomplete) answers at https://t.co/idiHOnOQTs

[Mirco Ravanelli @mirco_ravanelli]

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).

https://t.co/DR5B3IOsHj @MILAMontreal