Theoretical only? [LWN.net]

Theoretical only?

Posted Jan 10, 2013 10:37 UTC (Thu) by NAR (subscriber, #1313)
Parent article: A few words about Simon 0.4.0

I don't know how useful was at that time, but IBM OS/2 Warp included speech recognition some 17 years ago on the desktop. I think that was definitely outside laboratory research.

to post comments

Theoretical only?

Posted Jan 10, 2013 13:23 UTC (Thu) by sorpigal (subscriber, #36106) [Link] (5 responses)

Speech control, not speech recognition. ViaVoice and similar products have allowed speech to text for a long time, with varying levels of accuracy, and while there have been some attempts to use this ability to invoke commands there hasn't been a great deal of effort put in to making something that's good for more than "run this program"--certainly not something that also runs on Linux and is open source.

Theoretical only?

Posted Jan 10, 2013 14:11 UTC (Thu) by n8willis (subscriber, #43041) [Link]

I still have an old RPM copy of ViaVoice for Linux, from Mandrake 8.0. The dictation worked okay after a dedicated marathon of training, but I could never get XVoice to do anything useful ~12 years ago ...or whenever it was.... Shortly thereafter, IBM sold ViaVoice off to some other proprietary speech software company, and there was never another update.

Nate

Theoretical only?

Posted Jan 10, 2013 21:36 UTC (Thu) by man_ls (guest, #15091) [Link] (3 responses)

My Mac did this circa 1996 with PlainTalk. You spoke one of several commands, the computer recognized which one and run it. I never could make it work reliably but I'm not a native English speaker. Frankly it looked cool at the start but really sucked a lot.

It is a pity that Free software is still at this level when puny mobile phones can recognize with great accuracy the name of a street and take you to it. Granted, they send a signature to Google servers and get back the result, but a desktop machine should be comparable. Still, progress is welcome.

Theoretical only?

Posted Jan 11, 2013 9:00 UTC (Fri) by keeperofdakeys (guest, #82635) [Link] (1 responses)

Remote speech recognition works well because you can have a massive database of samples, so you don't have to do much work. Local speech recognition doesn't have this, so you have to get by with a much smaller subset of that data.

Theoretical only?

Posted Jan 17, 2013 14:10 UTC (Thu) by redden0t8 (guest, #72783) [Link]

To elaborate on what you said, the Terms of Service for Google, Siri, etc... make me think that they actually keep incoming voice searches as samples. They can even automatically guess whether they were successful or not based on whether you re-send a similar query.

Theoretical only?

Posted Jan 17, 2013 14:15 UTC (Thu) by redden0t8 (guest, #72783) [Link]

As a native English speaker, I found it remarkably reliable.

That being said, I never found it *useful*. Unless you can perform complex actions à la Siri, it's faster just to use your mouse+keyboard. This kind of relegates simple spoken commands to only being useful as an accessibility aid or in niche situations (ie operating the computer when your hands are otherwise occupied).