Speech Recognition Copilot for the A2A B-17

A Living Legend
N602AC
Senior Airman
Posts: 112
Joined: 19 Jun 2012, 17:50

Re: Speech Recognition Copilot for the A2A B-17

Post by N602AC »

I'm just wondering, is it more difficult to write the DLL using Google voice, or more difficult using Microsoft SAPI?
Image

lequinne
Airman
Posts: 36
Joined: 06 Feb 2011, 16:25

Re: Speech Recognition Copilot for the A2A B-17

Post by lequinne »

Far more difficult using the Google API. The issue is this:

Google's engine is newer, and the back end of it and internal workings are much better, but at present the actual "API" - the interface - is almost non-existent. That is, basically all you can do is give it an audio stream with speech and get text back. You can't control what it hears or how it tells you what it heard. For instance, if you say "Open the cowl flaps", Google may hear it correctly, or it may hear "Open the couch flaps" (as it often does for me). After all, 'couch' is a much more likely word than 'cowl' in normal conversation.

With SAPI (which has been around for years), you have far greater control. For instance you can tell it exactly what wordings to recognize, so it will try to fit everything it hears into those sentences. So SAPI has its issues, but it will never ever think you said "Open the couch flaps". Then to make it even simpler it will just give me a few numbers instead of actual text (actually it gives me the text too, that's what I display at the top of the screen). So if you say "Gear up", what my program actually processes is, say, the number "35" (meaning gear) and the number "0" meaning (in the case of the gear) "up". You can say "Can you please retract the landing gear" or you can say "wheels up, please", and my DLL will still just get the same two numbers and handle them the same way.

With Google, I basically get a dump of raw text, which I then have to process to implement an expandable list of substitutions, say, "cowl" for "couch", "gear" for "dear", etc. Then I must parse the text further to extract the actual information I need, so that "retract the landing gear" produces the same effect as "gear up".

To summarize, SAPI has a less advanced speech algorithm at its core (hence it's much worse in "dictation mode"), but it is optimized to process exactly this sort of set of limited-vocabulary commands. Google has better recognition at its core, but to date is optimized for open dictation (and very limited, google-defined commands like "search"), and has not made an interface available which has this sort of application in mind at all. When they do, or someone else makes one somehow, SR programmers all over will rejoice. In the meantime, manhandling Google's raw power into an application it is not remotely optimized for is an enjoyably challenging exercise, but like I said, it's going to take a bit of time. :wink:

Cheers,
Farley

P.S. Dan, have you tried 1.2.5? I fiddled with the optimization to try to keep it from recognizing "off" all the time. Is it giving you any help?

User avatar
Kilstorm
Technical Sergeant
Posts: 555
Joined: 08 Sep 2009, 22:00

Re: Speech Recognition Copilot for the A2A B-17

Post by Kilstorm »

This topic has been put to rest for awhile but the actual addon is still amazing and immersive. Not only that but I have been using nothing but taildraggers in Oculus Rift. Small crapped cockpits that are amazingly detailed thanks to A2A but decided to try the B17. OMG the feel of being in this spacious flight deck and then with the Oculus Rift mic, being able to just talk normal and have the crew hear you and response to your orders...total immersion. I then switched to night time and enjoyed the UV and glow in the dark lighting along with the weathered VC done by warbird...just amazing.
i9-13900K O/C | ASUS ROG Maximus Z790 Hero | 32GB DDR5 6000MHz Kingston FURY | RTX 4090 24GB | 2x SSD M.2 (2TB Samsung 990 PRO) 1x SSD (4TB Samsung 870 EVO) | Windows 11 Home | H20: HydroLux PRO:HardLine Tubing| 1000w PSU

User avatar
Lewis - A2A
A2A Lieutenant Colonel
Posts: 33284
Joined: 06 Nov 2004, 23:22
Location: Norfolk UK
Contact:

Re: Speech Recognition Copilot for the A2A B-17

Post by Lewis - A2A »

Good to see this bumped up, as you say Kilstorm its a fine thing to have up the top of the forums for others to see and use.

thanks,
Lewis
A2A Facebook for news live to your social media newsfeed
A2A Youtube because a video can say a thousand screenshots,..
A2A Simulations Twitter for news live to your social media newsfeed
A2A Simulations Community Discord for voice/text chat

User avatar
Kilstorm
Technical Sergeant
Posts: 555
Joined: 08 Sep 2009, 22:00

Re: Speech Recognition Copilot for the A2A B-17

Post by Kilstorm »

Could this same addon work with the Connie since it too has an inertia start?
i9-13900K O/C | ASUS ROG Maximus Z790 Hero | 32GB DDR5 6000MHz Kingston FURY | RTX 4090 24GB | 2x SSD M.2 (2TB Samsung 990 PRO) 1x SSD (4TB Samsung 870 EVO) | Windows 11 Home | H20: HydroLux PRO:HardLine Tubing| 1000w PSU

User avatar
Masterius
Senior Airman
Posts: 216
Joined: 30 May 2019, 14:36

Re: Speech Recognition Copilot for the A2A B-17

Post by Masterius »

This is simply fantastic!
Image

new reply

Return to “B-17 Flying Fortress”

Who is online

Users browsing this forum: No registered users and 8 guests