We found LALMs can answer ~50% of questions with SILENT input on the major benchmark.
We present
AudioMCQ: largest LALMs post-training dataset (571k samples)
&
Audio-Contribution-Aware SFT-RL post-training paradigm, achieving new SOTA
Paper: https://t.co/3yeIVZSd8K
#AudioAI
I'm super excited to announce that the MSR paper has been accepted at MMSP 2025, and the inaugural MSR challenge has been accepted as a grand challenge at ICASSP 2026! Thanks to Zheqi Dai, @QiuqiangK and @markplumbley for joint work on the paper, and @yaqubhai , Wanying Ge, Helin Wang and Prof. Kong's joint work on the challenge proposal. We'll share more information soon. Looking forward to advancing the next generation of solutions to the cocktail party problem!
📢Excited to announce the 1st workshop on 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗮𝗻𝗱 𝗣𝗿𝗼𝘁𝗲𝗰𝘁𝗶𝘃𝗲 𝗔𝗜 𝗼𝗻 𝗖𝗼𝗻𝘁𝗲𝗻𝘁 𝗖𝗿𝗲𝗮𝘁𝗶𝗼𝗻 (𝗚𝗲𝗻𝗣𝗿𝗼𝗖𝗖) at #NeurIPS2025 ! See you there🌴
🗓️Deadline: August 22nd, 2025 (23:59, AoE)
📷More info: https://t.co/bDxV2zFdSs
Why existing music source separation models are already so good (easily >10 dB SDR), but mixing engineers rarely use them? Our answer: They need music source **restoration**.
We introduce the music source restoration (MSR) task, which not only aims to separate the instruments, but also restore them to a state before any EQ, Compression, Reverb, Distortion and codec artifacts (e.g. MP3). We introduce the first dataset for this task, and found that the restoration task is much harder than separation.
Paper: https://t.co/94g8khpXXw
Model + Code: https://t.co/IoC2nPV2I7
Dataset: https://t.co/Eh6ZoZTZJo
HuggingFace Spaces: https://t.co/Sjg2sb93hP
Work done with Zheqi Dai, @markplumbley and @QiuqiangK . We are organizing a MSR challenge - if you are interested in participating, co-organizing or just learning more about this, please reach out to myself ([email protected]) or Prof. Qiuqiang Kong ([email protected])!
Looking forward to talking about Noise Network Plus at the Institute of Acoustics (@ioauk) London Branch hybrid meeting 6pm-7pm Wednesday 14 May 2025.
(Registration for in-person attendance closes 5pm Mon 12 May)
More below and at https://t.co/H0GSHubl44
Coming up 14 May 2025: 'London Branch hybrid meeting 'Noise Network Plus: A new interdisciplinary network to address noise pollution' presentation by Prof Mark Plumbley, University of Surrey (Centre for Vision Speech and Signal Processing) 19:00 - 19:00 AECOM, London, E1 8FA #acoustics #sound #noise #vibration https://t.co/sZUFNpvKJS
Coming up 14 May 2025: 'London Branch hybrid meeting 'Noise Network Plus: A new interdisciplinary network to address noise pollution' presentation by Prof Mark Plumbley, University of Surrey (Centre for Vision Speech and Signal Processing) 19:00 - 19:00 AECOM, London, E1 8FA #acoustics #sound #noise #vibration https://t.co/sZUFNpvKJS
JOB: Research Fellow in Generative Audio AI
Applications from under-represented groups encouraged, incl. women, people from Black, Asian and minority ethnic groups, & people with disabilities.
https://t.co/tU0zkFK0tH https://t.co/c2jQmWmB9v
@cvssp_research@PeopleCentredAI
SPS Journal at #ICASSP2025 today:
Apr 11: 2pm-3:30pm SAM-PJ1 (Poster 3G)
Selective-Memory Meta-Learning with Environment Representations for Sound Event Localization and Detection
Jinbo Hu, @Kalario_Cao, Ming Wu, @QiuqiangK, Feiran Yang, P, Jun Yang
AAM https://t.co/yz1w6Qc3dr
Talk at #ICASSP2025 this pm:
Apr 11: 2:30pm-2:45pm
A decade of DCASE: achievements, practices, evaluations and future challenges
@AnnamariaMsros@RSerizel@HeittolaToni@TuomasVirt P
AASP-SS6-L7: 50 years of Audio and Acoustic Signal Processing
Preprint: https://t.co/4vHuNDdtFY
On now at #ICASSP2025:
@Arshdee71825259 and Gabriel Bibbo presenting our demo:
Apr 11: 11:30am-1:00 pm
Personalized live sound recognition using efficient PANNs
Arshdeep Singh, @LiuHaohe, Gabriel Bibbó, Thomas Deacon, Mark D. Plumbley
Show and Tell Session II (Room: MR1.06)
Our paper at #ICASSP2025 today:
Yi Yuan et al, w @wang_wenwu: "Sound-VECaps: Improving Audio Generation With Visual Enhanced Captions"
Apr 10, 5:00pm-6:30pm, AASP-P22
Preprint/code/dataset/demos: https://t.co/vbVnw2VqST
// @cvssp_research@PeopleCentredAI
Our paper being presented at #ICASSP2025 today (Apr 8: 5:00pm-6:30pm)
FlowSep: Language-Queried Sound Separation with Rectified Flow Matching
Yi Yuan, Xubo Liu, @LiuHaohe, Mark D. Plumbley, @wang_wenwu
AASP-P4: Audio Source Separation I
More at: https://t.co/Ml9dRx2ayU
Delighted to be at the launch of Noise Network Plus last week, the £1.8 m project where engineers, policymakers, industry, social scientists & campaigners have joined forces to create quieter products, buildings & transport systems, aiming to cut noise over the next 10-15 years