📢 We are recruiting participants for the 1st European #MachineLearning BootCamp from July to October!
🔎 Are you aspiring to become a ML #Developer and get a job in the industry? This program is for you!
👉Learn more & Apply before June 20: goo.gle/ml-europe
Follow
Click to Follow TomKenter
Tom Kenter
@TomKenter
Research at Google UK on NLU and TTS.
ai.google/research/peopl
I also 
music: distrokid.com/hyperfollow/mo
music: distrokid.com/hyperfollow/moTom Kenter’s Tweets
Quote Tweet
I've been elected as a fellow of the International #Speech Communication Association (#ISCA).
Huge thanks to my colleagues & collaborators who have taught me so much & #Google which has given me opportunities to work on real-world problems.
isca-speech.org/iscapad/iscapa
1
3
OMG, I am stuck in a loop... I just keep on playing these two songs, one after the other, again and again, and I can't stop....
distrokid.com/hyperfollow/mo
�� RT if you're into heavy music��
#ObscureBandFromAmsterdam #MountTurmoil #rock
1
1
3
Show this thread
Friday afternoon... time for some rock & roll!
🎉💥🎸
This is me and a couple of friends of mine performing one of our own original tunes, live in a studio.
It Ain't Easy
Mount Turmoil — live
Video:
youtu.be/qNJhpzwrC_8
💥��💥RT if you like it💥��💥
1
2
Good news: the timeline has shifted so reviewers have more time for reviews. There will be emails sent out, but the new deadline for reviewing is ***Dec 27th.***
Quote Tweet
Replying to @ReviewAcl
Breaking news: The @aclmeeting program chairs have pushed out the ACL 2022 commitment date to Jan 15, thus allowing for a reviewing period of 11/29-12/27. Updated timeline at aclrollingreview.org/modified-timel. Emails will go out to AEs and reviewers.
3
15
46
Show this thread
Dear authors of ***ALL*** papers I have been reviewing lately...
Could you please increase the font size in your figures?!
You might want to use the size of the characters in main text as a reference.
🎊🎊 Thank you!! 🎊🎊
#NLProc #TTS #AcademicTwitter #AcademicChatter
2
5
Ha ha. Spot on, is what that is... ;-)
🎉�🎉
Quote Tweet
+100 to that! ;-)
Quote Tweet
Replying to @m__dehghani @Thom_Wolf and 4 others
Nice work!! I’m also a big proponent of “just fork the baseline and copy/paste code�. Adding too many abstractions is a common software engineering pitfall.
1
2
Of this is your topic, this is your ticket!
All these people are great, so go for it ;-)
Quote Tweet
PhD vacancy on an awesome topic, with two great supervisors!
Also there are very good benefits for PhD students in the Netherlands; and Radboud University in particular provides a very healthy and welcoming research environment.
Come join our team 
twitter.com/djoerd/status/…
twitter.com/djoerd/status/…1
Q: If you add kernel regularization to a Keras layer while using a custom training loop... will this do exactly nothing at all (if you don't also add model.losses to your overall loss)?
#AskingForAFriend
:-|
#TensorFlow #Keras #MachineLearning #NLProc
1
3
Mental note to self: if you add regularization to a network, you should actually add the extra loss term this introduces to your total loss or nothing will happen... Or you can just forget about it, and be mystified (frustrated) for weeks (like I was) :-/
#MachineLearning #NLProc
1
4
Schedule (in CET)
1300 Welcome
1330 Marzieh Fadaee
1355
1420
1445 Break
1515
1540
1605
1630 Break
1645
1705
1730 Panel “Transformers at Work�
More info zeta-alpha.com/registration-d
1
4
9
Show this thread
So exciting! 😃 The #igluContest has officially started!
If you are interested in interactive grounded language learning, definitely consider submitting!
We have two tracks, check them out below 👇
#NeurIPS2021
Quote Tweet
We are happy to announce that the submission system is up and running! So #igluContest has officially started:
Architect Task:
competitions.codalab.org/competitions/3
Builder Task: competitions.codalab.org/competitions/3
Please register and try it out!
#NeurIPS2021
4
21
Last week we released an evaluation set with particularly tricky lines for #TTS systems:
google.github.io/chive-prosody/
It was already used in #INTERSPEECH2020 and #INTERSPEECH2021 papers.
Please retweet if you are into evaluation and TTS! ;-)
#NLProc #GoogleAI #GoogleResearch
Quote Tweet
Nice news if you care about evaluating #TTS systems
Today we release an evaluation set for hard TTS cases:
google.github.io/chive-prosody/
I used in my #INTERSPEECH2020 paper, and it was already used in a couple of #INTERSPEECH2021 papers too.
#NLProc #GoogleAI #GoogleResearchShow this thread
3
7
18
We found that using a targeted dataset like this helped in developing our models. By making this dataset available, we hope to support the use of targeted and openly available test sets, and to make comparison between models more consistent and straightforward.
2
Show this thread
🎉 Nice news if you care about evaluating #TTS systems 🎉
Today we release an evaluation set for hard TTS cases:
google.github.io/chive-prosody/
I used in my #INTERSPEECH2020 paper, and it was already used in a couple of #INTERSPEECH2021 papers too.
#NLProc #GoogleAI #GoogleResearch
1
3
13
Show this thread
Credit where credit's due! 🎸🎉�
Quote Tweet
My mind is still blown. 
twitter.com/billboard/stat…
twitter.com/billboard/stat…Hard to believe it’s finally here but it's true. My debut album, Child of the State, is out NOW!!!
Listen loud. ayron.lnk.to/childofthestat
44
109
333
Funny thought: ML models are typically considered � if they can predict what humans do in a certain situation. But the reverse holds true too. I.e., ML researchers are often considered � if they can predict how a certain network architecture will perform on a particular task.
3
The review form for #ACL2021 had an "Ethics Review" section. I think it would be great if this would be included on all review forms from now on (@INTERSPEECH2021 for example) #NLProc #interspeech2021 #ethics #ethicalai #aiethics #responsibleai
1
2
New paper from our team:
Isaac Elias, Heiga Zen, Jonathan Shen, Yu Zhang, Ye Jia, RJ. Ryan, Yonghui Wu
"Parallel Tacotron 2: A Non-Autoregressive Neural TTS Model with Differentiable Duration Modeling"
Arxiv: arxiv.org/abs/2103.14574
Samples: google.github.io/tacotron/publi
2
32
135
Show this thread
New paper from our team:
Ye Jia, Heiga Zen, Jonathan Shen, Yu Zhang, Yonghui Wu
"PnG BERT: Augmented BERT on Phonemes and Graphemes for Neural TTS"
Arxiv: arxiv.org/abs/2103.15060
Samples: google.github.io/tacotron/publi
1
25
99
Show this thread
``Parallel Tacotron 2: A Non-Autoregressive Neural TTS Model with Differentiable Duration Modeling. (arXiv:2103.14574v1 [cs.SD]),'' Isaac Elias, Heiga Zen, Jonathan Shen, Yu Zhang, Jia Ye, RJ Ryan, Yonghui Wu, ift.tt/2PxQDeq
7
25
#SongOfTheWeek: Just Because by Mount Turmoil 🎆⚡�💥
Whenever we play this we always imagine a stadium full of people swaying to the beat and singing along at the top of their voices.
Just saying... ;-)
open.spotify.com/track/1hUWGVpn
music.apple.com/us/album/just-
I recorded a mini-album!! ;-)
It's available from:
Spotify: open.spotify.com/album/46DV0fsP
iTunes: music.apple.com/us/album/mount
Apple Music: music.apple.com/us/album/mount
Deezer: deezer.page.link/ZphqECL8cdj3u6
and a whole bunch of other streaming platforms.
Listen!
Tell me what you think!
Share ��
10
Ha ha. This thread...
(Kind of) smart and (rather) silly fun ;-)
Quote Tweet
1
New work with Ron Weiss, , , -- finally achieving what and I set out to do in 2016 before switching to spectrograms: direct waveform generation from characters. (1/7)
abs: arxiv.org/abs/2011.03568
samples: google.github.io/tacotron/publi
1
37
94
Show this thread
The team at have nicely summarised some of the TTS research from last month’s
1
6
12
Show this thread
Steenkolenengels (noun): if you English speak by just every Dutch word in the English to translate.
Quote Tweet
2
Highly recommended!! (pun impossible to avoid... ;-))
Quote Tweet
PhD Vacancy: Search and Recommendation Systems that Learn from their Users
I am looking for a PhD student to work on the next generation of ranking systems.
Fully funded position, great working environment, very friendly colleagues!
RT appreciated :)
ru.nl/english/workin
4
Woohoo! Hear, hear,! 🎊🎉
Quote Tweet
On Friday November 27th at 16:00 (CET), I will defend my PhD thesis titled:
Learning from User Interactions with Rankings: A Unification of the Field
While you should not judge a book by its cover, it is the only thing that I can share right now. 
(Design done by myself.)
(Design done by myself.)Show this thread
4
1
51
62
There are models to identify the language used in a piece of written text which can recognize 1000+ languages. What happens if you apply them to as many websites as possible? Turns out, you have to strengthen the classifier by quite a bit to get usable results. But when you do...
Quote Tweet
Language ID in the Wild: Unexpected Challenges on the Path to a Thousand-Language Web Text Corpus arxiv.org/abs/2010.14571
3
24
54
Show this thread
At #interspeech2020 and wondering about combining #NLU and TTS?
Have a look at "Improving the Prosody of RNN-based English TTS by Incorporating a BERT Model"
Main takeways:
- small BERT models work better than large ones (<-- !!)
- finetuning them helps
3
13