Athens, Greece
Joined February 2009
- Tweets359
- Following1K
- Followers291
- Likes4.1K
Our papers at #ICASSP2023 :
Speech-based emotion recognition with self-supervised models using attentive channel-wise correlations and label smoothing
Parameter-efficient transfer learning of pre-trained Transformer models for speaker verification using adapters
See you in Rhodes
Weakly supervised training on speaker recognition. We start with the original VoxCeleb videos, we keep only the audio and train a ResNet extractor with only video-level (not segment-level) annotation. We got 1.8% EER on the VoxCeleb extended test set.
arxiv.org/abs/2203.15436
Themos Stafylakis retweeted
Before Christmas, Olda, Ania and Lada went to Athens to continue investigation of NIST 2021 speaker recognition. The trip was supported by ESPERANTO, a Marie-Curie networking project . Their photo with Themos and Acropolis is simply lovely ;)
esperanto.univ-lemans.fr/en/…
The code from our @INTERSPEECH2021 paper "Speaker embeddings by modeling channel-wise correlations" (Omilia and Brno Univ. of Technology) can be found here. github.com/tstafylakis/Speak…
Excited that our paper "Speaker embeddings by modeling channel-wise correlations" has been accepted at #INTERSPEECH2021. arxiv.org/pdf/2104.02571.pdf
Themos Stafylakis retweeted
Check out our #BMVC2020 paper: “Seeing wake words: Audio-visual Keyword Spotting” with @afourast, @themosst, @SamuelAlbanie, and Andrew Zisserman!
Q&A sessions today at 10:00-11:00 and 16:00-17:00 UK BST: paper_042_P1_id_0043.
Project page: robots.ox.ac.uk/~vgg/researc…
Happy to share our paper "Seeing wake words: Audio-visual Keyword Spotting", a collaboration with the VGG team at University of Oxford. The paper will be presented at BMVC-2020. robots.ox.ac.uk/~vgg/researc…
Exciting times for Omilia techcrunch.com/2020/05/04/om…
Happy to share our latest work on speaker diarization. "Probabilistic embeddings for speaker diarization", Brno Univ. of Technology and Omilia Conversational Intelligence. arxiv.org/abs/2004.04096
In Omilia we are happy that our paper on end-to-end neural architectures for spoken language understanding has been accepted as oral at #ICASSP2020! See you in Barcelona! arxiv.org/abs/1910.10599
Check our new paper on Voice Biometrics (collaboration with Brno University of Technology). We show how to train speaker embedding extractors with self-supervision, enabling the use of unlabelled utterances in training. arxiv.org/abs/1904.03486
Check our new AudioVisual approach with CTC/Attention loss. We attain 63.5% WER on the challenging LRS2 using visual-only information, and 7.0% WER with early AudioVisual fusion, compared to 8.3% WER with audio-only (accepted at @SLT2018 ). arxiv.org/abs/1810.00108
Themos Stafylakis retweeted
Nips review:"You did not compare to *random paper that was put onto arxiv a month plus after nips deadline -potentially during review* and without this baseline we do not think the paper has merit. Hence no novelty."
The most unethical and toxic thing I have seen.
ACs, your turn.
Check our ECCV-2018 paper on Query-by-Text keyword spotting using only visual information on videos from BBC. xitter.gq/arxiv_cscv/status/1021…
This Post is from a suspended account. Learn more
Our new architecture for extracting word embeddings for lipreading, which attains 11.9% WER on excerpts from BBC-TV. arxiv.org/abs/1710.11201
A short survey about peoples' views on machine Lipreading and AudioVisual Speech Recognition. goo.gl/forms/LwJrfUTQdifGdVc…
Make a 3D model of your face here. From our lab at University of Nottingham. cs.nott.ac.uk/~psxasj/3dme