Probably both. All 3DGS needs is the ability to open it up and play the format. The dBpoweramp audio convertor I linked to has an encoder plugin and it'll compress a non-compressed original voice file (probably a WAV) to the spx format. If you don't use any spoken dialogue or very little then it probably isn't something to use, however for projects with ALOT of spoken dialogue this is what it would be best used for. See the Facade example I gave in the first post. I remember when it was first released it was a massive 800MB download. I just recently looked at it again after a recent story on it posted at Gamasutra and as of version 1.1 everything had been compressed to the spx format and it reduced the size to 167MB (13XMB for the Mac version since an additional bit of info. isn't required for the Mac version).

While it is true that a lot of 3DGS projects aren't using a lot of spoken dialogue, it's also true that a lot aren't using bones animation, shaders, or multiplayer, yet we have access to those features should we need them. I'd love to be able to use Vorbis for music (which I am), Speex for speech, and Theora for any video files I include. All three formats are stored inside an OGG file, so it's a matter of including the other format (or two if Theora ever makes it in). I just hope the audio "hiccup" that others have cited can be eliminated. The entire process is pointless if a split second delay occurs every single time a new audio file is loaded. The Open Al and FMOD plugins eliminate this hiccup, but would it exist if 3DGS natively supported Speex?