[dsockwell]

So someone asked in IRC if there were any alternative TTS voices for FSO, because Microsoft Sam sounds like he's stuck in the '90s. And he is.

So. I propose two things.

One is the inclusion of Festival ( http://www.cstr.ed.ac.uk/projects/festival/morevoices.html ) for text-to-speech, since it's actually a modern engine. I haven't looked into the details but it's the type of thing that should be easily called by FSO.

The second is that people (like myself) who don't have time to voice-act a campaign might release Festival voices trained with a minimum of two hours of recorded talking. This should open up a variety of virtual voice actors for whatever campaign that doesn't have the resources for real ones.

The cool thing about Festival voices is that if you're careful in training them, you can give them different moods. You'll see on the demo page I linked that one voice can be happy or angry. One voice model could release a number of voices - for instance, bored, calm, tense, and DIVEDIVEDIVE HIT YOUR BURNERS PILOT

Edit: Coders may find this interesting - http://www.cstr.ed.ac.uk/projects/festival/manual/festival_28.html#SEC132

Edit: And it's X11 licensed - http://www.cstr.ed.ac.uk/projects/festival/manual/festival_2.html

[z64555]

Looks interesting. The voices are smoother speaking than the ol' Sam and much easier on the ears, but it's still obvious TTS. 

[dsockwell]

True. These are the boring voices from a british linguistics lab. Specialized voices for freespace might be jazzed up somewhat.

[Luis Dias]

I have a problem listening to these things: they are always interrupted by one or two harsh quick sounds. Anyone has any idea why?

[dsockwell]

which voice profile did you use, and what text did you enter?

[Colonol Dekker]

Anna (HTS 2011) 

Oooh, you are so handsome Dekker!

[chief1983]

I've been wanting a cross-platform TTS engine dropped in for some time so we could get it working on *nix too.  That's one I've had my eye on.

[Herra Tohtori]

Having different characters use different voices would be quite excellent too. Possibly as specified by the mission creator.

[chief1983]

If mods could include their own voicesets, I'm sure that would be beneficial.

[Colonol Dekker]

Would it not be better to  have a (for instance) 10 voice pack, like MediaVPs, that's optional, each voice pack (depending on the engine) is a few hundred meg right? Rather than each mod presenting it's own dozen or so voice sets (easily a GB+ additional unnecessary data) when they can change the pitch and speed on a communal core selection? (Maybe via "modname_TTS_cast".tbm?)

Ivona TTS is quite good, in a beta stage right now. But i digress :/

[MatthTheGeek]

Your request is equivalent to ask that the MVPs include the Raynor and the Karuna because they're widely used ships.

That's not what the MVPs are for. And as long as trained voicesets are freely available - and will potentially evolve, due to the whole training thing- there is absolutely no point in putting them in such an inadapted and fixed place as the MVPs.

[Colonol Dekker]

 


I didn't say MVP's, I meant a TTS vp containing the core syntax files for the various voices. Any behavioural
updates would be in the kilobyte range.

[dsockwell]

why not make individual voices separate from mods, or mods unto themselves? that way you can say that X mod depends on Y and Z voices, so there's no duplication.

[dsockwell]

also those 190MB voices are a lot more complex than i imagine freespace voices would be - they're compiled to be as complete as possible, given a ****load of recording. A vanity voice is not going to have any ****loads of recording, it will be at most 12 hours of the same person.

[dsockwell]

at a few kilobits

[chief1983]

I didn't say don't release a basic set, I'm just saying that instead of solely hardcoding it to a few voices, make it something that a mod could supplement.

[dsockwell]

http://festvox.org/cmu_arctic/cmuarctic.data

prompt list for prospective voice models

[logomancer]

Bear in mind that the TTS engine used to make Anna is not part of Festival proper, but is a separate program that uses Festival for text analysis. The TTS engine is found here, but some parts of it may be incompatible with the license for the FS2 codebase. The two engines that I know are part of Festival either a) sounds much like the MS speech API, or b) has large voice files and sounds like someone spliced up wav files to make the voice (which it kinda does).

If the HMM engine can be used, keep in mind that it takes the better part of 2 or 3 days to actually make a voice; the compiling will take on the order of 12 to 36 hours.

[Tomo]

IANAL, but the HTS licence sounds ok to me - it's a BSD licence, which boils down to:
"Do what you will, but don't claim this code is yours and don't claim we endorse your product"

So I don't think that's a barrier, just the fun and games of actually using it and training it.

As chief said, getting something cross-platform is certainly a worthwhile project!

[jr2]

dsock, use "modify" button on your post, double-posting is frowned upon here.