Forums » Suggestions

Voice-chat to Text

Dec 11, 2014 Pizzasgood link
This is a pretty low priority suggestion.

One of the things I dislike about voice chat is that if I step away from the computer for a bit, I miss stuff. One way that could be corrected would be to have optional voice to text support linked into the voice chat system. This would be on the receiving end, not the transmitting end. If I enabled it, it would continue playing people's voice chats back to me as usual, but would also transcribe them into text and dump that into the appropriate channel of my chat log (with some kind of flag to signify that it was voice, not typed).

It would also be useful for the occasional deaf player; yes he could just tell his friends to please type everything out, but the hands-free communication that voice chat offers is very useful when you need to convey information during a fight. With a voice-to-text system, his friends could continue taking advantage of VC without leaving him completely out of the loop.

Of course, voice-to-text isn't perfect and it would make some errors. I think it would be worth it though. But yeah, low priority.
Dec 11, 2014 tinbot link
I would think that this can be accomplished client side, by installing voice to text software on your computer, and having it capture your computer's audio output.

This could be an feature that could be developed, just preferably not developed by guild software.

I would be, and am, surprised, that there is not already some generic application that does this.
Dec 12, 2014 csgno1 link
What I do now is run mumble on my phone while I'm playing VO on my laptop. If I need to walk away I just slip the phone into my pocket so I can keep with the conversation. I hope the VC client of the future is not necessarily tied to the game client, I like them separate.
Dec 12, 2014 joylessjoker link
I'm fully in favor of this as one of those occasional deaf players to play VO, but unfortunately voice to text isn't easy to implement. For it to be fully effective and functional, it needs to be calibrated to individuals' voices, and the database needs to be filled with vocabulary appropriate to VO. Otherwise you get ridiculous, incorrect results like on youtube when you turn on automatic captions. Even the best voice to text software on market, Dragon NaturallySpeaking, has those restrictions. We need major strides in voice recognition technology before we have anything decent that doesn't require calibration.
Dec 12, 2014 incarnate link
Google basically has the best in the world, and it requires a machine learning AI that has tens of thousands of profiles (or more, millions now?) that are tuned to different types of speakers, with each individual's voice/recognition history also tracked and profiled, and then response data sent back to their servers (massively parallel supercomputer, basically), for analysis that's seeded with the profiled user's history. Ie, they can throw infinite computing power at the issue, and a lot of relatively simple but sensible AI on a massively parallel level, and it still occasionally has issues.

Anyway, I dunno, it's a pretty computationally intensive feature request, that will probably not work super-well in the best case scenario. But it might still be useful even if it's mediocre. A friend of mine in that world said it's pretty easy to get like.. 80-90% accuracy, but after that it's super difficult.

So, yeah, neat idea, but I don't see us pouring resources into it. If someone writes an amazing, friendly-license, open-source voice recognition thingie, then maybe, but even then there would be some meaningful integration time to make it all work.