Neural Network Recreating Voice (Kind of)

Discussion in 'General Off-Topic Chat' started by Attacker3, Oct 30, 2016.

  1. Attacker3
    OP

    Attacker3 GBAtemp Regular

    Member
    211
    297
    Mar 24, 2015
    Canada
    Underground, in my mother's basement.


    Turn your volume down or you will get a headache, otherwise it's a fantastic video showing how a nerual network can teach itself how to do things.
     
    Engert and Zero72463 like this.
  2. Zero72463

    Zero72463 GBAtemp Maniac

    Member
    1,221
    581
    Jun 27, 2016
    United States
    Cool and creepy.
     
  3. Attacker3
    OP

    Attacker3 GBAtemp Regular

    Member
    211
    297
    Mar 24, 2015
    Canada
    Underground, in my mother's basement.
    Very. Later in the video it ends up sounding like a baby crying, which I find really interesting.
     
  4. VinLark

    VinLark This machine kills bourgeois sentimentality.

    Member
    4,065
    4,672
    Jun 11, 2016
    Trinidad and Tobago
    4chan and other wonders of the internet
    Woah, that's really interesting
     
  5. TotalInsanity4

    TotalInsanity4 GBAtemp Supreme Overlord

    Member
    6,927
    6,909
    Dec 1, 2014
    United States
    Under a rock
    Yikes, it actually sounds like a baby babbling

    It'd be interesting if someone decided to "raise" a neutral network in real time like a child, especially if it were designed to develop itself further. You could theoretically teach technology morals
     
  6. Engert

    Engert I love me

    Member
    887
    481
    Jan 21, 2012
    United States
    Taxachusetts
    Very interesting.
    Thanks for the vid.
     
  7. FAST6191

    FAST6191 Techromancer

    pip Reporter
    23,193
    8,944
    Nov 21, 2005
    I am of two minds
    1) I would love real life narrated a la bastion, though you could probably do that with conventional phoneme capture.
    2) This sounds like it could do a wicked take on the snoopy voiceover style thing where you have sounds and distorted voices tailored to the situation. I was too lazy to see what hardware was used for this but what was there sounded within the realms of shared computing or something I could leave my little home/work cluster working on for a couple of weeks (no different to long 3d or video renders in ways that matter to me). Wonder what the law would say there as most sampling laws are based upon rather different concepts. However if I can skip straight to full on narration then I might not see this as much, maybe have some things narrated and the rest this.

    I should also like to see voices sampled with something like this to do... a capella style songs if the term can be abused that much.

    Interesting times we live in.
     
    TotalInsanity4 likes this.