Home Forums Soap Box speech recognition in games Reply To: speech recognition in games


My concern for a stable in-game solution, putting aside the closed nature of console hardware for a moment (but which you’d have to contend with as I expect more and more XB360/PS3 to use the feature), is what impact this feature would have on game performance and *possibly* what trade offs would have to be implemented to maintain the gameplay experience (FPS/bots/eye candy).[/quote:7724cc9287] From a development point of view this is certainly the big one. The large system resource requirements would currently be a real problem.

Moreover, I would imagine that there would be quite a vast amount of work to be done when you’re contemplating localisation… [/quote:7724cc9287] Not as much as you might think, provided that the VR was used in a structured way within the game. For example, in a military game you would have standard commands such as “attack”, “enemy 12-o-clock”, “take cover”, “suppressing fire” etc. From a programming point of view these are just command 1, command 2, command 3 etc. The program knows that “advance” is command 1 and when it detects it it tells the game what to do. If you replace the English dictionary/database with a Italian database the Italian equivalent of attack would still be command 1* and the program would deal with it in exactly the same way.

But it would be nice to -say- play BF2 SP with bots with which you can interact through voice recognition…[/quote:7724cc9287] If we ignore the system resource issue then this becomes quite easy. Point your aiming cross-hair at an enemy machine gun nest and say/shout “suppressing fire” and the program would tell your AI team mates what map co-ordinate you were looking at and what command number “suppressing fire” was. They would then use their AI to find the nearest cover and go into their suppressing fire routine (pop off a few shots, duck down rinse and repeat). The enemy AI would know that they are on the receiving end of suppressing fire and would act accordingly (duck down, attempt to spot your team, attempt to return fire but in a less coordinated fashion than if they were not under fire).

As there are only so many orders you can give in a game ( eg “Shoot the feckin fecker” rather than an order like “4 pints of guinness mate” ), the challenge would be to quickly allow for fuzzy comparison between what is said, and what it might be similar to ( if anything ).[/quote:7724cc9287] This is actually an AI problem, not a voice recognition issue. In a military game you could get around this by making it part of the game play that orders must be given clearly. The military train to use standard commands so that everyone clearly understands. “Enemy 12-o-clock” is a clear command warning of enemy and its location “Look out, Krauts” would be a poor command and if the AI didn’t understand it it would be reasonable to have them hit the dirt but not know what to do until a clear command was given.

*Actually the Italian equivalent of “attack” is “surrender” but that is a different issue ;)