13 replies to this topic

[Breunor]

This website contains the following statement:
Festival is a general multi-lingual speech synthesis system developed at CSTR. It offers a full text to speech system with various APIs, as well an environment for development and research of speech synthesis techniques. It is written in C++ with a Scheme-based command interpreter for general control.

Using the proposed Sable markup language and Festival, we could create an interface for converting any text file into speech output. These tools are Opensource IIRC. Why would anybody care? Consider this:

At the end of the month, we currently get a summary of how things went in the game-score, change in income, etc. Current plans lean towards more info about what happened, perhaps in a narrative form. Now take this narrative, convert it to speech, and have a 3D newscaster model sitting behind a desk like CNN Headline News, mouthing the words as they're spoken. Screenshots of completed research pop up on the screen next to the person as they talk about each item, etc. An added bonus, if people were so inclined, would be to make spoof commercials that could be inserted before and after the news, and randomly selected each time. I'm thinking along the lines of the commercials that were in Total Recall and Robocop. People could make them and submit them for others to download, dump them into a mod folder for random selection.

I really like this idea, but I have no clue when it comes to Linux programming. So I'm tossing this out there for anybody who might be interested in it. It will surely take some time to develop, but so will this game, right?

[GreatGold]

That looks amazingly sweet. I think we should add something along these lines to a "to do" list (hint hint, like the one your developing?? ). If its technically feasable, it would do great as you propose, but could also add some more avenues of improvement for v2.0+. Love it.

Think of using this technique for briefing before a mission (if we ever feel the want/need to add one), your about to land the assault troops, arm them in the arming screen, and then right before you disembark: a screen of someone, perhaps the pilot or ops officer back at base, pops up and gives you a short brief to the tune of general environment, time, general location, suspected alien race, suspected alien ship., or whatever have you. I think that would add something nice, not to mention the intel could always be off (they tell you sectiods suspected, turns out to be mutons) as a function of skill level played at. Lots of fun ideas for future Xenocide versions.

Like I said, an excellent thought - we could really do some fun stuff with something like this...anything that isnt scripted could be narrated. Very sweet.

Gold

[Cubik]

Could be extremly nice, IF the voices can sound anyway near natural. This seems to be the problem with most text-phaser programs.

As you (Breunor) pointed out in another post, it could also be used for the agents voices in battles. This would hopefully eliminate the need of voice actors, and would provide us with amazing possibilitys to customize the voices regarding sex, age, morale and more.

[Breunor]

Some of the speech synthesizers out there sound pretty good, the key is setting them to pronounce words correctly. Xenocide would probably have to be spelled Zeeno-side for the app to read it properly, but there are features that can fix that. If only we had the hooked on phonics for computers program!

[Breunor]

Here's a sample of what I envisioned for the newscast at the end of the month: Ananova

There's no recorded voices, this is all generated text to speech output. Down the road we could possibly work something like this into the game.

[Micah]

Very nice! How easily could we make this talking ana?

[Deimos]

I like it. We could work with the artificiality of the presenter and say it's X-net's AI presenting updated info at the end of each month.

[Cubik]

Suprisingly good, I really think that we should try and work into the v.1 release. The voice still has some glitches here and there but it sounds much better than any speech synthesis system I have ever heard.

[Breunor]

It's my understanding that Ananova is not available for anything other than her intended/current use. :crying:

I doubt we could get access to the coding for any of it, but the parts to consider include: The 3D model with plenty of joints for animating of course, some type of timing code that matches certain facial animations to specific phonic output, and hardest of all a competent speech engine. Current research on speech engines seems to be focused on phonic interpretation of text, but Ana is by far the best I've heard at pronouncing this clearly. The timing code is unique to that project I believe, it would be quite a hassle to recreate it. There are several speech engines out there, they vary in quality of output quite a bit. Good ones that I used when training the blind did a good job, but there was no assumption that it was a real person like Ana is supposed to be. Meaning, the mechanic nature will stand out more. I'll see what I have around here, maybe I can find some info as to how we could use the engine.

I think this would be quite a monster to get ready for version 1, without pulling people from other areas, which would slow everything down. And if the output isn't to people's liking, it becomes a lot of wasted effort. On the positive side, the speech engines can be upgraded in a modular format, I was able to swap out engines on the fly. So as improvements come around, we could upgrade quickly. I'll take a look at what resources are out there for general use and get back to you all on it. I'd really like to see an interface that uses something like Festival, and we could feed it Sable-tagged text to test.

[red knight]

Most games are pretty constrained in time, and generally those text to speach systems are resource drainers... That means that featuring that kind of system will increase maybe to unrasonable limits the entry level system.... It is always a trade off, almost nothing is imposible (as i said in another thread) the important thing is if you are using the resources efficiently... For example: I would ask: Do i need a realtime speeach system for Xenocide it i already know what i want to say and can generate that offline???? Maybe it is because i am Computing Oriented, but there is always a better way to do it without sacrificing quality...

Greetings
Red Knight

[Breunor]

I've done some more research on this, and there's no feasible way to include nice sounding speech into the game using opensource software. Ananova's speech system can be bought relatively cheap, but it's not opensource, and takes 500mb of space to store its phenomes. The opensource versions don't sound real enough even to consider for a robot/computer voice in the game IMO.

If and when an opensource, small footprint engine that sounds good becomes available, and if our end of month output can be put into a tagged XML format (that Sable markup can be applied to), we can look at modifying things, but as RK just said, it's a relative waste of time right now.

[Breunor]

Since this is version 1+, I'm moving it to the labs.

Edited by Breunor, 26 December 2003 - 01:33 PM.

[CptJackSparrow]

Would be nice. If we had a usable text2speech engine
I could even imagine talking soldiers in battlescape
because we could have thousands of comments and
thousands of voices and it wouldn't be so annoying.
But as red knight said, we have to consider resources,
especially in battlescape...

For a monthly debriefing we could record
a voice actor and then let the program do
some "cut&paste" where needed (like in Madden,
FIFA or similar games).

[Puasonen]

I think those text to speech programs still doesn't sound anywhere near real. They could be used in v1+ if those programs will be devolped a lot more..

edit:
Not to mention we should find one FREE

Edited by Nyyperoid, 30 December 2003 - 04:09 AM.