Julius Richter

Truly a pleasure to be involved in building such a versatile sound event detection model. Huge shout-out to my great collaborators: @apoorv2904 @mhnt1580 @BowenShi20 @bigpon517 👏

Fun example of PE-A-Frame confidently predicting “a man speaking” and “a man coughing”, but being much less confident about “a man predicting the future” 😄 (Though @SchmidhuberAI may have been right back then) 📄 Paper: tinyurl.com/pe-av-paper 💻 Code: tinyurl.com/pe-av-code

@ymas0315 Haha nice! And was thinking about a DJ mixer. Maybe the auto-translation messed up 😆

@JuliusRichter13 Vitamix for smoothies😅

ブラックフライデーに論文通ったので調子に乗ってお高めのミキサーを買ってしまった

🙏 Many thanks to my supervisor Timo Gerkmann for his invaluable guidance, to the reviewers Shinji Watanabe and Simon Leglaive for their insightful feedback, and to commission members Sören Laue and Jianwei Zhang. Grateful to all collaborators who made this journey so rewarding!

🎓 I’m thrilled to share that I successfully defended my PhD on generative speech enhancement at the University of Hamburg! My work explored diffusion-based and audio-visual generative models for robust speech enhancement and audio restoration. 👉 tinyurl.com/thesis-richter

@adriiiarizav2 Code

Nicee! I have a lots of code invite for Sora 2, FOLLOW ME and comment here “CODE” and I will send by X chat 😊🤘🏽

@ArxivSound Reference [6] is not my work.

Naisong Zhou, Saisamarth Rajesh Phaye, Milos Cernak, Tijana Stojkovic, Andy Pearce, Andrea Cavallaro, Andy Harper, "Shortcut Flow Matching for Speech Enhancement: Step-Invariant flows via single stage training," arxiv.org/abs/2509.21522

@forthshinji I am curious about the result for generative SE …

草 (音声強調を繰り返しかけると無になる)

@serrjoa Absolutely. I also have speech enhancement/restoration in mind. Replacing corrupted areas in the spectrogram using CLAP similarity on speech.

Let's do this for audio, no?

@wuyusongwys Amazing! Thanks for answering the questions my brother asked you 😊 I couldn’t make it to ICML this time

We are presenting our poster soon at West Exhibition Hall B2-B3 W-502, at 3:30-4:30pm! Check it out online: icml.cc/virtual/2025/p…

@BenUFO @notyumiko1000 What's the one at 1:55:00? Pure fire 🔥

it's up! on.soundcloud.com/E5kWDqX3RVKxGi… @notyumiko1000

@wuyusongwys Great contribution!

It’s been a thrilling journey building FLAM! 🚀 Super proud of what we achieved open‑vocabulary audio event detection using calibrated frame‑wise modeling. FLAM will be presented at ICML 2025, come check it out! 📄 Paper: arxiv.org/abs/2505.05393 🎧 Demo: flam-model.github.io

@justin_salamon @AdobeResearch @MIT @Mila_Quebec @wuyusongwys @tsirigoc @Kotentorothy @huangcza @AaronCourville @urinieto @pseetharaman Great work! Will you release the code/checkpoints?

I think we finally cracked it? FLAM can detect *any* sound via text prompts arXiv (ICML'25): arxiv.org/abs/2505.05335… demos: flam-model.github.io @AdobeResearch+@MIT+@Mila_Quebec led by @wuyusongwys w/@tsirigoc @Kotentorothy @huangcza @AaronCourville @urinieto @pseetharaman

@ArxivSound Great work!

``FLAM: Frame-Wise Language-Audio Modeling,'' Yusong Wu, Christos Tsirigotis, Ke Chen, Cheng-Zhi Anna Huang, Aaron Courville, Oriol Nieto, Prem Seetharaman, Justin Salamon, ift.tt/vfyP8jR

@wataru9871 👏

ｴｯﾎ ｴｯﾎ ｴｯﾎ ｴｯﾎ 残響を保持した音声復元ができるって伝えなきゃ ｴｯﾎ ｴｯﾎ ｴｯﾎ ｴｯﾎ 残響の制御もできるって伝えなきゃ ｴｯﾎ みんなに伝えなきゃ paper: arxiv.org/abs/2505.05077 demo: google.github.io/df-conformer/r…

``Normalize Everything: A Preconditioned Magnitude-Preserving Architecture for Diffusion-Based Speech Enhancement,'' Julius Richter, Danilo de Oliveira, Timo Gerkmann, ift.tt/0WZoJm7

Check out the slides here: shorturl.at/TUK5G. Please note that the PDF is 36 MB in size due to its audio and video content, and it is best viewed using Acrobat Reader.

Join me tomorrow for a webinar on "Generative Audio Restoration in Multimodal Applications"! I'll introduce the tractable Schrödinger bridge and discuss the differences between flow matching and score-based methods. signalprocessingsociety.org/blog/sps-webin…

Will present our paper, "Investigating Training Objectives for Generative Speech Enhancement" at #ICASSP2025! 🗓 Wed, 5:00-6:30 PM (AASP-P12) 🔊 Discussing diffusion bridges (Schrödinger bridge) & connections to score-based models—let’s chat!