Text to Speech : how we could overcome the voiced protagonist.

[Uroak]

Hi.

 

I think that many mod makers are facing this issue, when creating new quests for instance. Despite the fact that there are already many lines of dialogue to choose from, it is not possible to be precise enough when writing dialogues. And no one knows all those lines, so searching must be a burden.

 

But it is 2016 (huzzah !). I think we have the technology to solve this issue. And not by making the player silent again. So my idea was the following : use all the dialogue lines of the SS as a training base to be able to generate new ones. The ones we want. The training base may be sufficient.

 

People remembering the text to speech synthesis of google translate may think that the result will be horrible. And they are right.

Also, people interested in IT stuff may think of neural networks, and how they allow to solve many problems, without understanding anything about the problem. And they are right too.

 

So here is an article about a recent text to speech system (using neural networks) : link

Also, those not familiar with scientific articles should instead read this (and listen to the examples given to be convinced) : link

 

I think we have the scientific tools to generate new lines of dialogue, imitating nicely the voice of the SS, either male or female. Isn't is nice ? I should also mention two tools found here in the nexus mod list that could help in the process :

- Unfuzzer, to extract audio files.

- Dialogue File Renamer, to easily associate an audio file with the attached string.

I think you guys must know a few more tools that can be helpful.

 

 

I do not have the scientific knowledge to understand the scientific material enough. I also do not have the coding experience to put all this in practice. I am mainly writing down my thoughts here, for the curious reader or the interested modder. Let just say that I am fascinated by what can already be done as of today.

[werr92]

That's a very interest ing article you've provided. Shall read it.

 

On the subj. If we're speaking about the ways other modders use, me, for example. I have a fully custom 100 lines of dialogue for both male and fem characters voiced by voice actors. Don't see a big problem here, oblivion had a low voice variety and sometimes even happened for one npc to speak multiple voices. But still it was a great game.

[EmissaryOfWind]

I mostly skimmed the article, but the examples provided feel very close to human speech. Correct me if I'm wrong, though, but the software that generated those examples doesn't seem to be available to the public, and would require processing power that goes beyond the capabilities of most people's PCs, and I don't think whoever developped it would let us use it for free. At that rate, it becomes quickly a more viable option to pay a real human voice actor. But it would be interesting for game companies, in order to generate for example the characters calling the player by the name you put in, rather than recording thousands of possible names. It could upgrade it from small "easter egg" that happens once at the start of the game to a real feature that carries on throughout the whole game.

[zrovektor]

I mostly skimmed the article, but the examples provided feel very close to human speech. Correct me if I'm wrong, though, but the software that generated those examples doesn't seem to be available to the public, and would require processing power that goes beyond the capabilities of most people's PCs, and I don't think whoever developped it would let us use it for free. At that rate, it becomes quickly a more viable option to pay a real human voice actor. But it would be interesting for game companies, in order to generate for example the characters calling the player by the name you put in, rather than recording thousands of possible names. It could upgrade it from small "easter egg" that happens once at the start of the game to a real feature that carries on throughout the whole game.

The software is being developed by google. No, it is not available to the public. It's not even ready for actual production use. When it will be, if ever, is anyone's guess.

 

https://techcrunch.com/2016/09/09/googles-wavenet-uses-neural-nets-to-generate-eerily-convincing-speech-and-music/

 

http://www.theverge.com/2016/9/9/12860866/google-deepmind-wavenet-ai-text-to-speech-synthesis

"WaveNet isn’t going to be showing up in something like Google Assistant right now — its step-by-step processing system requires a comparatively huge amount of computing power."

Edited September 16, 2016 by zrovektor

[onelunglee]

Have been experimenting this with myself. Posted the idea in my thread a few days ago.

 

https://forums.nexusmods.com/index.php?/topic/4856350-injecting-myself-into-fo4-without-breaking-immersion-at-all/