TTLG|Jukebox|Thief|Bioshock|System Shock|Deus Ex|Mobile
Results 1 to 12 of 12

Thread: AI voice replicator

  1. #1
    Registered: Aug 2009

    AI voice replicator

    Not sure if you guys are following the other forum. There's a new voice replication AI, and it's incredible, it can mimic voices perfectly. This can be a game changer for FM's, as you can now replicate Garrett's voice. A few examples from the thread:


    Keeper trainer

    Mechanist 1

    Noble 2 (this one is brilliant)

    Servant 1

  2. #2
    Registered: May 2005
    Location: France
    Wow that's pretty great!
    The Garrett voice is missing some personality, but it's pretty damn close. The others are all great though!

  3. #3
    Registered: Oct 2001
    Location: The Crippled Burrick
    Guess that's me sacked.

  4. #4
    Registered: Apr 2011
    i avoided posting in your other duplicate threads about this—why spoil your fun? but now youve come and put it on my doorstep, as it were, i feel i have to finally respond:

    no thanks.

    a creative human work is important: as a thing in itself; as a component of the growth of a person's own skills; and as a vital necessity for sustaining communities, of both creators and audiences.

    the designed purpose of these ai tech startups to destroy all of that, to devalue creative labour by mass-producing an ersatz copy. when you thoughtlessly adopt their tools, you are declaring that you also have no respect for the value of the work. we are here as a community making creative works, missions, that are confections of level design work, 3d modelling work, 2d texture work, music and sound design work, and voice acting work. proposing an automatic mechanical substitute to replace even one of those parts displays not only a deep disrespect for the work that goes into each of them, but also a profound ignorance of how the creative aspects of all of those parts interoperate together to produce a great mission; they're not isolated components that you can pull out and plug something else of vaguely the right shape in.

    imitating Garretts voice perfectly was never the point of having voice acting in a mission. sure, its great when a voice actor can do a natural-sounding imitation of one of the original voices; it helps with a sense of continuity of the character. but that continuity is a bonus; its far more important that your voice actor understands the character, and the context of the lines, and produces a performance that fits into the entire creative structure that youre assembling when making a mission. take, for example, Yandros's Benny: his voice sounds very little like Stephen Russell's Benny; but Yandros knows the character well and produces a lively and entertaining performance that is Benny—just, if youd like a metaphor, a different incarnation of Benny than in the original games.

    lastly, not because its the most important factor, but the least: please listen to voice acting more. maybe try doing some voice acting yourself, even just as a temporary placeholder recording in a mission, to understand what goes into it. it is clear from your praise of the samples you posted that you have not developed a good ear for performance. the samples do not sound at all like Garrett, except insofar as the distribution of frequencies approximates the timbre of his voice. they mostly sound like a decent text to speech algorithm reading text: like a very bored voice actor churning out lines without understanding any context or having any interest in the output. i hope thats not the kind of performance anyone here actually wants in their missions.

  5. #5
    Registered: Apr 2001
    Location: Lost in the BSP...
    Yeah.. I've thought about a bit more, and as vfig has put so eloquently, I'm just not into that no matter how good the "representation" is. If FM authors want to use it for their FMs, then I say that's their choice, but I wouldn't like to play those as I would know (and glaringly so) that it's not Garrett (or SR). It's a replica. It would just feel wrong.

    One of the main things for me is the "heart" that goes into a lot of those VA roles. We have some extremely talented taffers here that really make some of those roles shine. One of the best examples I've seen of that lately was in vfig's Making a Profit. Seemed like every corner I turned there was another couple of AIs 'conversing' and they were downright unique and entertaining; several were hilarious. I just don't see an AI voice replicator being able to reproduce that, not for a long time anyway. Personally, I'd rather have someone who sounds quite a bit different if it means having really quality voice performances available to us. These folks do it for free, and we're certainly the richer for it, as creators and players.

  6. #6
    Registered: Jan 2001
    Location: Constantly losing tug o'war
    I also prefer to have humans doing the voice acting. I'll use a speech generator for placeholder lines while the building is in progess, as it makes it easier for me to change the script as I see fit. Once I'm happy with that (usually after beta testing has started) I'll then ask for humans to do their thing.

    I can, however, sympathise when someone wants a minor bit of speech to go with a regular AI that will act as normal from that point, but even so, we can usually find someone who's close enough, and if not, a different voice can lead to a different approach being taken by the author, which can benefit the creative process.

  7. #7
    Registered: Apr 2001
    Location: Lost in the BSP...
    They're okay-sounding for the deadpan delivery parts (like Garrett's monologues) but to me they just sound too "dead" for parts that require feeling and emotion. Maybe it's just me.

  8. #8
    Registered: Mar 2012
    Location: Moscow, Russia
    How long until the AI can create fan missions?

  9. #9
    Registered: Mar 2012
    Location: Moscow, Russia
    Isn't it possible to refine these sounds to make them more alive? First of all, any AI makes a good reference, a base for final processing. It looks a bit boring as it is, although it's ideal if you want to use different non-living creatures: necromancers, anonymous souls from Rocksbourg, undead, golems, robots, mechanist slave servants, etc.

  10. #10
    Registered: Apr 2011
    Quote Originally Posted by Soul Tear View Post
    First of all, any AI makes a good reference, a base for final processing.
    i disagree even with this, though not because of ai specifically:

    you should try to record your own scratch recordings for all voicework in your mission. it helps your brain to engage with it as speech and not written word, and you will discover by doing this places where your wording feels unnatural, or where another characters response doesnt flow right, and similar problems. so you come out of the process with both the scratch recordings ready for beta use, and knowledge of how to polish your dialog for the final draft.

    it doesnt take much time or effort either: you dont have to care about levels much, or noise reduction, or silence trimming, a scratch recording can have bad audio quality. and you dont have to try to do a great performance to benefit from this effect, just do a performance; even if your scratch recording ends up being a bad performance that you definitely dont want, you can just take notes on that but let the scratch stay, for efficiency; try not to do retakes unless e.g. a coughing fit prevents you getting the whole line out. you can barrel through Record, Save, Record, Save and get scratch lines done super fast while still getting the now-i-know-how-to-polish-the-writing-better benefits.

    this effect is a good example of what you lose out on by throwing the task over to text to speech, or “ai”. your own brain is already an excellent intelligence that improves every time you engage it with something. thats a fantastic ability that you should try to benefit from whenever you can.

  11. #11
    Registered: Jan 2006
    Location: On the tip of your tongue.
    AI aside, you can achieve what you're suggesting vfig by just reading your script aloud.

  12. #12
    Registered: Sep 2008
    Location: Deutschland/Germany
    Quote Originally Posted by Hit Deity View Post
    They [...] just sound too "dead" for parts that require feeling and emotion.
    Yes, and I think that is the flaw a virtual machine won't be able to solve soon. All you can do is changing loudness and speed of the WAV.

    Nevertheless if there is a quick translation required for an upcoming FM those voice engines have been useful for some simple dialogues instead of recruiting "real" voices in a hurry. I tried ...

    ... but this voice replicator only seems to speak english so far, isn't it?
    Last edited by zappenduster; 19th Feb 2023 at 04:48.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts