Talky Morrowind

[Amenophis]

I think they use native windows voice synthetizers.
details here: http://forums.bethsoft.com/topic/977902 ... e-dialogs/

[Chris]

The way it looks to me, given how things are animated and the variable delays before the speech starts depending on the dialog length, it uses code injection to take control of certain aspects of menus and dialogs, so that it can make NPCs idle while menus are open, and when new dialog is displayed it feeds the text to an external text-to-speech synth for the NPC (probably writing the results to a temp wav file, then when it's ready it makes the NPC 'say' that file).

[Capostrophic]

I'll now be sorta cap'n obvious.

If I understood Amenophis correctly, he's wrong, Talky Morrowind uses external voice synthesizers with a choice of Windows's native text2speech voice(s) if it is (they are) present, of course.

On the first run speech is of course not pre-recorded, it's generated real-time with some modifiers applied to genders and races (pitch, speed etc.) configurable in the program. Using of recorded files is configurable too, IIRC.

The problem is, voice "actors" are ridiculously wrong with toponyms and names pronunciation.

[MikeMXellers]

I tried Talky Morrowind and it's pretty neat, it's obviously robotic but I think it's really cool as a feature, the voices are generated on the fly based on installed voice fonts with the ability to tweak things like pitch speed etc for voice fonts and a myriad of other options like pausing time or zooming into npc faces, I don't think the game caches generated voices which I think is a good thing. I hope this feature ends up in the OpenMW codebase, sure it's rough around the edges like a durian but a feature like this would be a great way to drum up interest in OpenMW, as long as it's optional and off by default that is. I'd love to see a croaky dunmer voicefont, or a smooth, accented Khajiit voicefont.

[johndh]

sure it's rough around the edges like a durian

Hard to get into, loved by few, and banned on public transit? Even if there's a really good voice library that would make it sound natural (which I doubt) and that is customizable enough to make Dunmer, Khajiit, and Argonian voices, it still seems like such a niche idea and so hard to get right.

[Zobator]

sure it's rough around the edges like a durian

Hard to get into, loved by few, and banned on public transit? Even if there's a really good voice library that would make it sound natural (which I doubt) and that is customizable enough to make Dunmer, Khajiit, and Argonian voices, it still seems like such a niche idea and so hard to get right.

But if done right, it'll be mindblowing

[Nivim]

I think it would be hilarious to build one of those deep, recurrent neural networks everyone is using lately; find a few thousand recordings that have the voices you want, pair those recordings with accurate transcripts and the right race/gender, and then feed those to the network until the wonderful mess generated sounds about right.

Of course, this isn't something that would ever work on the fly; neural networks usually take a lot of processing power, and need to run a long time both to learn and to output.

Edit: If you're unfamiliar with what these networks tend to do, see DeepForger for one that applies effects to images, RoboRosewater and it's Thread for one that tries to generate M:TG cards, and this old Hexahedria post on musical neural networks.

[GreyFox]

This seems like something one could get away with in a collaborative project more than a commercial one. After all, we do not have to fear the wrath of the console kiddies with things like this. The direction TES went with voice acting is responsible for a lot of loss in the RPG area. Voice acting is expensive and clumsy, meaning once you recorded it there is little you can do if you notice that different dialogues might be better story wise. I would like to see this as a feature of openmw(with a disable option of course), but I doubt someone is willing to play around with it. The cost in terms of time it would take is in questionable relation to the benefits. Immersion is not necessarily benefited by robotic voices and there is the legal stuff as mentioned before.

A quick research yesterday about the most advanced opensource synthesizers led me to MARY. Here is an online Demo:

http://mary.dfki.de:59125/

I tried this sentence with various voices and tweaking the parameters, the results were quite good actually.

"Rest well this night, for tomorrow you sail for the Kingdom of Daggerfall."

(try to do that with one of the German voices ... hilarious)

[Okulo]

TVoice acting is expensive and clumsy, meaning once you recorded it there is little you can do if you notice that different dialogues might be better story wise.

Not to mention modifications. I remember finding an NPC in Vivec telling her stories about her travels around Vvardenfell. Only later I found out that this NPC was part of a mod. Mods are able to plug in way better into the world if you leave voice acting out.

[Chris]

I remember finding an NPC in Vivec telling her stories about her travels around Vvardenfell. Only later I found out that this NPC was part of a mod.

There's mods for Skyrim that add NPCs that you'd be hard pressed to tell apart from the vanilla game, too. Interesting NPCs being one example, as well as large scale quest mods like Moonpath to Elsweyr, Helgen Reborn, and Falskaar. Conversely, it's not as simple to just add new fitting NPCs to Morrowind as some make it out to be... it's not hard for me to be able to pick out lines and NPCs added by LGNPC, for example, due to things like spelling errors, bad punctuation, out of place dialog, etc. While it's undeniably harder to make new NPCs fit in with voice acting, it definitely adds to the atmosphere to have NPCs that can actually talk to you.