Bence Halpern

👀 slideshare.net/slideshow/symp…

NeurIPS 2025(世界最高峰のAIの国際会議)で注目を集めた827論文(Oral + Spotlight)を国別にマップしたaiworld作成の地図 ビジネス業界中心の米国 vs 学術界中心の中国 日本も結構頑張ってるように見えますが、分類的にはRest of the world.

ホノルルで開催された「Symposium on Speech & Behavior Informatics」にて、音声合成の歴史から最新のニューラル生成モデルまでを個人的な視点で外観する講演をさせていただきました。 企画していただいた皆様、ご来場いただいた皆様、ありがとうございました！ speakerdeck.com/ktokuda/thirty…

finally finished my 🦑 reviews i don’t want to review 10 papers for a conference again

Please do not use UTMOS score as "MOS."

See you in the Netherlands 💺

I'm wondering is there anyone who reads papers (IEEE-style) on 100% magnification or less (excluding people who print out)? Even with my glasses on, I find 130% much more comfortable to read.

@unilightwf At the time of the publication it should definitely be runnable, especially if it’s stated as a contribution. Otherwise, papers with good open source code get cited more, so it’s not like this is not unpenalised but the current system.

Many papers nowadays use "we open-sourced our code!" as a contribution, but what if the code is just not runnable, and the authors are just not responding to all the issues on GitHub? Ran into so many rubbish open-source repos recently, getting really pissed...

Save the date for the 16th Advances in Quantitative Laryngology, Voice, and Speech Research (AQL) conference! 🌍 Groningen, the Netherlands ⏰ June 24 - 27th, 2025 📥 Abstract submission opens November 18th, 2024

Check out the article @tejarebernik and I wrote on why kinematic speech measures are so important when studying those with motor speech disorders! Available in Dutch and English below ⬇️

@i15fujimura1t 伝説のシャチもいる？

TD研サメマゲドン合宿

I feel there is an uncanny valley where a paper is just intimidating enough to say “it looks novel” but actually it is just very poor writing. I’ve been asked to revise such papers and I’m so torn because I feel making it better might just reduce the chance of acceptance 😅

Dear speech researchers, ICASSP'25 is looking for a reviewer in the speech and language areas. Please nominate yourself via docs.google.com/forms/d/1wtydY…! Note that this form is only for the speech and language areas.

We need your help to make ICASSP 2025 a success! If you have previously been a reviewer & have not received an invite (check spam), please contact the TPC chairs: tpc@2025.ieeeicassp.org We apologize for any confusion caused by the multiple email invites due to a portal error.

I ended up with using ResearchRabbit and tagging negative papers in the search so I don’t look at things twice. Not an ideal solution but allowed me to go lot more deeply with my review that I usually can.

Speech tech/linguistics people! What is a comprehensive literature review tool that you have positive experience with? 😅 Haven’t found anything that has all the features that I want 😕

To my academic colleagues and network: @nvidia offers an Academic Grant program with up to 32,000 A100 hours for research projects in Generative AI and LLM, as well as simulation, physics, fluid dynamics and quantum computing. The deadline for the current round of applications is at end of Q3 (September), and the next round starts in Q4. nvidia.com/en-us/industri…

To DCASE Task2 people We conducted new experiments on this year’s dataset🫡 arxiv.org/abs/2409.09332

@OxfordWave @Cambridge_Uni @Selwyn1882 @CamPhonLab Congratulations! 🎉 Can’t wait to read the final thesis 👀

An ecstatic end to four years of hard work - Linda Gerlach celebrating after completing her PhD viva at the University of Cambridge @Cambridge_Uni . She completes the Oxford Wave Research PhD studentship with Selwyn College @Selwyn1882 ! Pictured at the @CamPhonLab & Selwyn.

Many inspiring works during Interspeech so I rly want to write a pathological TTS/VC paper again but the evaluation of it and answering to the reviewers' (completely justifiable) comments of why this task is different in all aspect compared to a typical TTS/VC task is pure hell.