I was wondering..is the talk in the game(for example: Fifa,Nba), is that an AI? or is that only IFs..if the ball in the basket do that..for example.. thanks..

that's not AI. they do it like you said. if a certain action happens, play this audio file.

in my opinion, announcers in sports games are just as annoying as announcers in real life. adding a bit of AI to it would make them more enjoyable, but probably a complete waste of cpu and memory usage.

That code probably still counts as "AI" (after all, in the games world, even a bot that makes decisions with the function "rand(2)" is considered AI). It's just not very interesting or sophisticated AI.

Any form of intelligence would ruin the simulation of the commentator.

The complexity of game VO range from production rules/finite state machines/expert systems implementations to complex pattern matching systems (using decision trees, Marvok chains, etc).

Depending on the game, there is definitely a range of techniques. One company was going to publish their method in the last AI Wisdom, but the paper was pulled at the people up the chain considered it too valuable to share (or patentable, I don't remember the specifics).

Generally for "AI" audio in games you put audio hooks into different parts of the system. In an FPS, for example, you'll put hooks into: entity damaged, entity reloading, group behavior about to start, player damaged, player reloading, player taking cover, entity taking cover, etc.

If you place the hooks correctly you can get voice callouts before the action actually happens. It's a really easy way to trick players into thinking the AI is smart: "wow, that one dude said 'take cover' then everyone took cover"; in reality the AI had already decided that everyone should take cover so it told the audio system to play the callout.

In sports games you do the same thing. You sprinkle audio hooks all over the place. The audio system will then have a layer to process all the hooks. Audio designers attach various sound banks to each callout and tune the percentages for the individual wavs in each bank. They can do whatever meta-processing of the audio cues they want: no more than X voice callouts at a time, certain callouts are once per level only, etc. Basically, the audio system is spammed by the hooks and it gates the actual playing of sound in response to those hooks.

-me

There are a few reasons why most audio commentation in sports games aren't very smart. The main one is that the audio is pretty much pre-recorded. There's a long series of pre-recorded comments that get mixed and matched around based on the situation. So, the commentators can't say anything other than what is already stored internally. If there were good text-to-speech implementations built in, then that's completely different altogether. With good text-to-speech, you can then generate comments in real-time as seen fit. However, this comes with alot of added overhead, not to mention memory requirement. It's not impossible though, if they manage to finally perfect text-to-speech to the point where you can also control the mood of the commentator, then that would just make things perfect. Of course, I'm not saying that pre-recorded comments are any better though, in that department. We can probably name a few sports casters in real life that sounds the same all the time.

Side note:

The other problem with text to speech is localization. Even if we nail english, that doesn't necessarily mean we have solid TTS for all locales.

Maybe we will get there some day, but I don't see it happening any time soon.

Quote:

Original post by AN_D_K Any form of intelligence would ruin the simulation of the commentator.

This is sooo true :D
But the commentator is a finite state machine, isnt it? That makes it an AI.

Quote:

Original post by Gagyi But the commentator is a finite state machine, isnt it? That makes it an AI.

Actually, A finite state machine is not AI in itself. It can be used to create AI though. It's like saying a brick is a house, just because it can be used to build one.