voice to text/text to voice

#21 OK, I've played with the SDK a little and here are my findings with relation to games:

* Hardware requirements are rather hefty. TTS (Text To Speech) requires 14.5Mb RAM, and recommends 32Mb. SR (Speech Recognition) requires over 25Mb, and recommends 128Mb.

* SR is slow and requires a good amount of CPU time. My machine (350MHz PII, 64Mb RAM) takes a good minute to fully recognise a small paragraph (three-five sentences). I have a feeling it gets faster with training, but I wouldn't make any "attack on my signal" commands in a game :)

* SR can be horribly inaccurate. Getting a little frustrated with the system, I said (clearly) "this is stupid," which the computer identified as "barstool student." Yes, it gets better with training... but bear in mind that during a game, the player is hyped up and probably can't concentrate on speaking as clearly as is needed, and the background noises of the game will interfere. ESPECIALLY if the microphone picks up other players' speech. Could be a nasty feedback loop :)

* TTS is ok; doesn't require too much memory or CPU. However, it's a bit robotic (much better than in the Amiga days though). I couldn't find any documentation on creating new voices, so you're stuck with two male voices and one female. The pitch and speed can be changed, and words can be emphasised. (But not specific syllables, sadly. That removes a lot of possible speech artistry. :( ) However, there is the option to create WAV files of the speech, so a bunch could be recorded ahead of time and played back to suit.

As far as SR goes, it may be better to hold off until hardware acceleration is available. The TTS was disappointing, given the lack of options. (The old Amiga had more customisability!)

On top of that, the documentation is like DirectX's, but worse :)

Ahh well...

Grant

#22 Makes me wonder what MS Game Voice is using.....
I used MS Game Voice for a long time in Tribes and Tribes2 for deploying, droping, choosing load-outs.... basicly everything accept movement, firing, jumping and jetting.

#23 I assume Game Voice is a technology where you can give commands via voice?

I think the SR problem is greatly simplified if it only needs to recognise a few words. I guess it's the difference between having to compare sound against a dictionary of 20 words, and a full English language dictionary.

I definately know from observation that my CPU usage hovers at around 20% and takes only a second or two when expecting one of a few words, but maxes out at 100% and takes ages when doing the full SR job.

Grant

#24 If you want real time voice over net TeamSpeak is what I have found to be the best as far as low cpu usage and low ram usage for gamers . Actually you can set the amount of ram it uses.

http://www.goteamspeak.com/index.php?page=downloads

I believe this is the correct addy for the download . Although it isnt a speech to text type , it is a speech to speech . seems I read in the forums somewhere that there is a SDK to implement it into the game set it for team 1 and team 2 . Hope this helps you in finding what your looking for...

#25 Not sure.... if this is a record or not. It's gotta be close if not.

Okay think I got it right, it's 1632 days, two hours, and three minutes.

#26 Rofl.