Written by
Nickolay Shmyrev
on
Interspeech 2020 Thursday
Horay, this year I made an effort to review all Interspech paper, it
never happened for me for quite some years. Speaker recognition, emotion
recognition, ASR for language learning, transformers and semi-supervised
learning, all mixed in my head now.
Thursday is a bit more relaxed than before. To list papers which got my interest:
Thu-2-7-8 Improving on-device speaker verification using federated learning with privacy
Not much serious or exciting, just an important research direction looking on idea of differential privacy
Thu-3-3-3 Hide and Speak: Towards Deep Neural Networks for Speech Steganography
Another important privacy topic which is going to be more relevant it seems
Thu-3-10-9 Conformer: Convolution-augmented Transformer for Speech Recognition
Conformer demonstrates really great results. ESPNet got really great improvements implementing this architecture.
Overall, really cool conference. I would love to hear some more comments
and points about it, please share your thougths on Reddit or Telegram.