Papers of the day   All papers

Speech2Face: Learning the Face Behind a Voice

Comments

May 25 2019 👩‍💻 DynamicWebPaige

🗣️→👫 "How much can we infer about the way a person looks from the way they speak? In this paper, we study the task of reconstructing a facial image of a person from a short audio recording of that person speaking." We live in the future. https://arxiv.org/pdf/1905.09773v1.pdf @MITCSAIL @MIT https://t.co/K4m9dlDzqv
28 replies, 586 likes


May 26 2019 Jeff Clune

Whoah
2 replies, 73 likes


May 25 2019 Dorian Taylor

okay i am normally not moved by ML dog-and-pony shows but this is something
4 replies, 57 likes


May 26 2019 William Isaac

Who is demanding this kind of technology? Who benefits from deploying it in our communities? (h/t to @mdekstrand for flagging this)
3 replies, 44 likes


May 25 2019 cameron tonkinwise

it is just as easy to reverse engineer the type of person who is on the Institutional Ethics Review Board at MIT approving this research - and their lack of education
2 replies, 25 likes


May 25 2019 Michael Ekstrand

seven people thought this was a reasonable thing to do and a good use of human and computational resources. why? of all the things to try to do with AI, why pick this one?
2 replies, 21 likes


May 26 2019 Arthur Charpentier 🌻

I might understand that it could be fun to map different spaces (from voice to picture), but still... I think I don't like where it's going....
3 replies, 17 likes


May 26 2019 Jeremy Kahn

I mean, what could go wrong
0 replies, 13 likes


May 26 2019 Edouard Harris

2 years ago, I thought this sort of thing *might* be possible in 10 years
0 replies, 13 likes


May 29 2019 bakedtapes

“How much can we infer about a person’s looks from the way they speak? In this paper, we study the task of reconstructing a facial image of a person from a short audio recording of that person speaking” https://arxiv.org/abs/1905.09773v1 https://t.co/mCnO7Eer11
0 replies, 10 likes


May 31 2019 Herman Saksono

This neural network AI creates a person’s face based on their speech. My question is how to exclude a person’s accent when predicting their face? If we give a speech of an asian man with a thick Boston accent, will the algorithm render an asian man? https://arxiv.org/abs/1905.09773v1 https://t.co/Ut2oRYI0AO
0 replies, 9 likes


May 25 2019 Abeba Birhane

WHY??? I struggle to find one good reason why or how this can be helpful, unless of course, the aim is to amplify harmful stereotypes. https://twitter.com/DynamicWebPaige/status/1132293559216795649
4 replies, 8 likes


May 25 2019 Jonathan Korman

Do not worry about the technologies we are inventing. Plan for them. Start ensuring that our institutions can be trusted with them. https://twitter.com/dynamicwebpaige/status/1132293559216795649?s=21
1 replies, 5 likes


May 26 2019 Giorgio Patrini

Generative models are opening solutions to new inverse generative problems, e.g. face -> voice 👇 My favourite: image seen by subject -> fMRI brain activity, "mind reading" https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1006633 Claim: these demos are surprising/controversial when they invert causal direction
2 replies, 4 likes


May 25 2019 Timi Ajiboye

Lmfao man, but we dey here for Naija, no light.
1 replies, 4 likes


May 26 2019 rixavie

http://arxiv.org/abs/1905.09773v1 https://t.co/RahPV0hmAq
1 replies, 4 likes


May 25 2019 Dawn Ahukanna

@Abebab - another one for the collection. I usually say, “There are no bad ideas”. But I’m channeling Jeff Goldblum character meme for Ian Malcolm from Jurassic park - “Your ... were so preoccupied with whether or not they could, that they didn’t stop to think if they should”. https://t.co/xMPZNaHar9
0 replies, 4 likes


May 25 2019 Amanda Silver

Wow... this is incredible. I woke up dreaming that @Radiolab should do an episode on how technology can help visually impaired people. It would make such a great episode. So much to talk about and a great forum to learn!
0 replies, 4 likes


May 28 2019 Michael Merrifield

Holy cow! I had to check that this wasn’t an April 1st submission. It isn’t. (https://arxiv.org/pdf/1905.09773v1.pdf) https://t.co/0ZRqvHqsuf
0 replies, 3 likes


May 25 2019 Dan Kaminsky

This process ends up a lot more ... generic ... than anyone is really expecting. H/t @doriantaylor
2 replies, 3 likes


May 28 2019 Alexander Kruel

In another example of the uncanny abilities of artificial intuition, this machine learning model can predict skin tone, age, and facial features simply by listening to a recording of that person speaking. https://arxiv.org/pdf/1905.09773v1.pdf https://t.co/EU6MAbSo3B
1 replies, 3 likes


May 25 2019 Jeffrey Moro

I think one of my greatest challenges as a teacher of digital studies is going to be getting the STEM students who take my classes to see how dangerous these technologies are https://twitter.com/dynamicwebpaige/status/1132293559216795649
1 replies, 3 likes


May 25 2019 Travis Hoppe 🖤✖️◼️

Y'all thought that AI profiling was bad before 👇
0 replies, 2 likes


May 28 2019 Luca

In today's edition of "Let's encode stereotypes into ML models"
0 replies, 1 likes


May 26 2019 Jamie Stantonian

@Timcast If you want something even more dystopian check this: "How much can we infer about a person’s looks from the way they speak?" https://arxiv.org/pdf/1905.09773v1.pdf
0 replies, 1 likes


May 25 2019 tante

Language codes social status, class and habitus which in turn correlates strongly with looks. What else is new?
0 replies, 1 likes


May 26 2019 Ronald Richman

Wow...
0 replies, 1 likes


May 29 2019 Pawel Pachniewski

Speech2Face: Learning the Face Behind a Voice "We have demonstrated that our method can predict plausible faces with the facial attributes consistent with those of real images." https://arxiv.org/pdf/1905.09773v1.pdf https://t.co/ff0zCO6MFa
1 replies, 0 likes


May 29 2019 Matt Hanson ▙❘▟▂▁▄▎

Voice Face Correlation! Research showing reconstruction of a person's face through an audio capture #neuralnetwork #AI #futures #spooky ⇢ Paper: https://arxiv.org/pdf/1905.09773v1.pdf
1 replies, 0 likes


Content