Pinned Tweet
I'm incredibly thrilled to share that our project on competitive robot table tennis, Ace, has been published today on the cover of @Nature
For 40+ years, building a robot that could rally with an elite human table tennis player at full speed was an unsolved problem. Sony AI's Ace research project set out to change that—and the results are now accepted for publication in @Nature and featured on the cover.
#ECCV in 2 weeks! Attending & into vision for sports, tracking, real-time perception, or event cameras? Check out our spotlight paper: “Event-based Gaze Control Systems for Real-time Spin Estimation in Professional Ball Games”, Fri Sep 11, 14:25 CEST (spotlight), 16:00 (poster)
Looking forward to discussing with you in Malmö
Paper: arxiv.org/abs/2606.26780
Demo: youtube.com/watch?v=nj4a0T24…
43
Excited to share that our “Event-based Gaze Control System for Accurate Real-time Spin Estimation in Professional Ball Games” has been accepted to #ECCV2026.
This technology is used in the table tennis robot that beats professional players.
Paper: arxiv.org/abs/2606.26780
1
1
139
The @SonyAI_global table tennis robot: ace.ai.sony/
Latest match results and videos against professional players: ai.sony/blog/what-happened-a…
56
Curious to see how much more competitive our table tennis robot Ace has become since the #Nature paper? Check out the latest match results against 9 professional players — including two-time Olympic medalist and 2017 World No. 5 Miu Hirano.
Since our @Nature paper, #Ace won against seven pro table tennis players, including two-time Olympic silver medalist Miu Hirano, and world No. 26 @KiharaMiyuu. Learn how improvements across #RL, hardware, #perception, and more helped close the gap: bit.ly/440oyPM
138
Attending CVPR and interested in physical AI, perception, event cameras, or table tennis? 🏓Stop by the Sony booth (#329) to meet Asude Aydin from my team! She will be presenting our recent #Nature paper on the first table tennis robot capable of beating professional players.🔥
1
1
130
大変分かりやすい解説動画をありがとうございます！
ソニーAIロボットは如何にして卓球エリートに勝ったのか 論文から読み解くAIの潜在能力 youtu.be/jKnthsVSUJs?si=Zi8F… @YouTubeより
1
127
Low latency perception, agile robot HW, and dynamic control enable Ace to return the ball even when it suddenly changes its flight trajectory at net contact. Check out our project page for more details.ace.ai.sony/
What happens when a ball hits the net? In table tennis, net contact creates unpredictable trajectories. For Ace, Sony AI's physical AI research system, these rare events were one of the hardest real-world conditions to address.
1
2
169
Check out the story behind our work on the first table tennis robot that compete with professional players under the official rules.
Incredibly proud of being a part of this project.
youtu.be/FrGq8ltb-_E?is=IRZw…
2
145
Our paper titled "SilentCipher: Deep Audio Watermarking" got accepted in @ISCAInterspeech 2024. Watermarking signal introduced by SilentCipher is inaudible even in challenging band-limited and silent regions, yet is robust to variety of distortions.
arxiv.org/pdf/2406.03822
We are presenting our text-query sound separation paper @iclr_conf on 3rd May. If you're around & interested in audio, SSL, multi-modal learning, CLIP, etc., please come to our booth at poster see soon 6, #27.
Happy to share that I will be presenting our sound separation paper at @iclr_conf in Kigali! 🥳
In this work, we proposed the first text-queried universal sound separation model that can be trained using only unlabeled noisy videos! 🤩
#ICLR2023
youtu.be/kOcvnr-GYBE
7
772
Happy to share that our paper on diffusion-based singing voice vocoder got accepted to #ICASSP2023.
Demo: t-naoya.github.io/hdm/
Excited to announce that 3 papers got accepted at #ICASSP2023 🥳.
Many thanks to wonderful co-authors.
19
1,606
Excited to announce that our work on text-queried sound separation got accepted to #ICLR2023 ! Many thanks to reviewers for the thoughtful and constructive feedback!
CLIPSep: Learning Text-queried Sound Separation with Noisy Unlabeled Videos
We propose a text-queriable sound separation model that can be learned w/o text-audio pair data.
Work is done with and first authored by @salu133445 during an internship @Sony.
arxiv.org/abs/2212.07065
CLIPSep: Learning Text-queried Sound Separation with Noisy Unlabeled Videos
We propose a text-queriable sound separation model that can be learned w/o text-audio pair data.
Work is done with and first authored by @salu133445 during an internship @Sony.
arxiv.org/abs/2212.07065
One-shot singing voice conversion that is robust against reverb effect and interference from accompaniment music.
Demo: t-naoya.github.io/rosvc/
Paper: arxiv.org/abs/2210.11096
Thanks @_akhaliq for sharing!
Music Transcription using diffusion model. First authored by @ravencheuk.
DiffRoll: Diffusion-based Generative Music Transcription with Unsupervised Pretraining Capability
deepai.org/publication/diffr…
by Kin Wai Cheuk et al. including @zuNaoya
#Classifier #ComputerScience
8
Thanks @_akhaliq for sharing our work!
We use multiple diffusion models at different sampling rates to progressively generate waveforms. By properly handling the anti-aliasing filter, the proposed method improves the quality of generated samples.
Demo: t-naoya.github.io/hdm/
In 30min @ISCAInterspeech, our work with Shrutina Agarwal and Sriram Ganapathy @tweet4sri will be presented. Please come to Wed-P-OS-6-2 if you are around and interested in our work.
arxiv.org/abs/2208.12410
To be presented at @ISCAInterspeech 2022!
Leveraging Symmetrical Convolutional Transformer Networks for Speech to Singing Voice Style Transfer
deepai.org/publication/lever…
by Shrutina Agarwal et al. including @zuNaoya
#NeuralNetwork #ComputerScience
Today! Please come to our poster session MLSP-46 @ieeeICASSP , Thu, 12 May, 22:00 - 22:45 Singapore Time (UTC +8).
Got accepted to #ICASSP2022. We show that adversarial attack can be used to alter the separation performance of multiple target models individually in both positive and negative directions.