Tips for Mac OS X Speech Recognition

[MacBytes]

http://www.macbytes.com/images/bytessig.gif (http://www.macbytes.com)

Category: Tips and How To's
Link: Tips for Mac OS X Speech Recognition (http://www.macbytes.com/link.php?sid=20060101143224)

Posted on MacBytes.com (http://www.macbytes.com)
Approved by Mudbug

[SPUY767]

Apple was one of the first manufacturers to ship speech recognition. There were add-ons with system 7, and System 8 had fully integrated speech recognition. Alas, speech recognition was one of the many instances where Apple did something first, but failed to capitalize because of poor marketing or timing. Go Figure.

[Verto]

Am I the only one who notices a significant slowdown when Voice Recognition is on? I am able to use the voice commands with little problem, but the benefit/coolness of it is offset by the fact that my computer slows down far too much for my liking.

EDIT: And I'm not sure how this article was worth mentioning, other than to give the Voice Recognition feature some attention, since the author basically just copies the built in "setup wizard".

[Fender2112]

Am I the only one who notices a significant slowdown when Voice Recognition is on? I am able to use the voice commands with little problem, but the benefit/coolness of it is offset by the fact that my computer slows down far too much for my liking.
I wish I could get Voice Commands to work reliably on my wife. When they do work, things REALLY slow down. :D

[shamino]

Apple was one of the first manufacturers to ship speech recognition. There were add-ons with system 7, and System 8 had fully integrated speech recognition. Alas, speech recognition was one of the many instances where Apple did something first, but failed to capitalize because of poor marketing or timing. Go Figure.
Depends on how you define "fully integrated" and how you define "first".

OS/2 version 4 (released in 1996) had built-in voice recognition. It worked on any Pentium-or-better processor with a 16-bit sound card. (And it worked on 486's as well, but too slowly to be useful.) This voice recognition included full speech-to-text dictation. A speech-enabled version of Netscape (where all visible links were speakable items) was bundled with the system.

Mac OS 8 was released in 1997 (as you say, speech was an add-on in earlier releases), and to this day it still does not support dictation.

[MoparShaha]

I remember using speech recognition in 1995 on System 7.5 on my friends Power Mac 7100. Hasn't changed much since then.

[redAPPLE]

speech recognition would be like when Apple would allow windows installation on Apple hardware.

first off, i don't know how long it would take to train the machine to recognize my voice.

training would probably take long. how about people, who does not speak the words clearly? slang?

it would be like a million software drivers, that the machine has to recognize etc. (in windows)

[diddy]

I wish I could get Voice Commands to work reliably on my wife. When they do work, things REALLY slow down. :D

Bwahahahaha! Good one. :) :D

[jobberwacky]

Another limitation of ’speakable items’ is that it’s US-English only. Its functionality has never been extended to other languages and not even to other ‘Englishes’. The hit rate for people talking the ‘wrong’ dialect variant or for individuals with a slight accent is too small to make it useable.

In its current form it’s not much more than a gimmick.

[shamino]

first off, i don't know how long it would take to train the machine to recognize my voice.

training would probably take long. how about people, who does not speak the words clearly? slang?

it would be like a million software drivers, that the machine has to recognize etc. (in windows)
I obviously couldn't say what a hypothetical Apple-dictation module would be like, but I can write about how these questions apply to OS/2 (the only speech system I have personal experience with.) Hopefully, any modern system would work at least as well.

On OS/2, the system works OK without training. Basic training (a 10-15 minute procedure where you read a sequence of keywords into the mic) greatly improves accuracy. Full training (about an hour) makes it very accurate.

When dictating dictation, you have the ability to correct its mistakes. If you do, accuracy improves. After a few weeks of usage, while correcting it, I found that it would rarely make mistakes. It's only problem (which is really part of its design) is that you must insert brief pauses between words when dictating. As for accents and dialects, training usually takes care of that issue. I don't think it supports foreign languages, but I never installed a non-US version of the system.

WRT software, all of the standard system controls (menus, buttons, etc) are automatically made into speakable items. And all system text-entry controls support dictation. So many apps require no modification whatsoever.

For apps that use custom controls, often only minimal changes are required. Netscape was speech-enabled by creating a "Links" menu containing the text of all visible links (using ALT text for image links.)

For apps where this is not possible, there are APIs where the program gives a list of keywords (up to 1000, I think) to the OS. When one is spoken, a message is be posted to the app. There are also APIs to let your app accept dictation when system text-entry controls are not used.

WRT driver software, nothing special was needed. Any audio-input device capable of 16-bit input would work. (At the time, there were still many 8-bit audio cards in use.)

[840quadra]

Depends on how you define "fully integrated" and how you define "first".

OS/2 version 4 (released in 1996) had built-in voice recognition. It worked on any Pentium-or-better processor with a 16-bit sound card. (And it worked on 486's as well, but too slowly to be useful.) This voice recognition included full speech-to-text dictation. A speech-enabled version of Netscape (where all visible links were speakable items) was bundled with the system.

Mac OS 8 was released in 1997 (as you say, speech was an add-on in earlier releases), and to this day it still does not support dictation.

Quadra and Centris AV Macintosh computers had PlainTalk back in 1993. It was not an add on software package, and was included with the system, but was dependent on the DSP. PlainTalk used the onboard AT&T DSP to both talk, and hear, and unlike later versions, didn't impact overall system performance as it was being driven by a separate processor.

It fared quite well (considering it's newness) in this article by Byte magazine.

http://www.byte.com/art/9401/sec9/art5.htm

[bousozoku]

Depends on how you define "fully integrated" and how you define "first".

OS/2 version 4 (released in 1996) had built-in voice recognition. It worked on any Pentium-or-better processor with a 16-bit sound card. (And it worked on 486's as well, but too slowly to be useful.) This voice recognition included full speech-to-text dictation. A speech-enabled version of Netscape (where all visible links were speakable items) was bundled with the system.

Mac OS 8 was released in 1997 (as you say, speech was an add-on in earlier releases), and to this day it still does not support dictation.

The difference is in training. PlainTalk never required you to take time to train the system. While IBM's technology was more comprehensive, I could not get ViaVoice to work reliably at all while I used PlainTalk on a daily basis. Having what I call a Japanglish accent, the IBM technology and training should work much better.

Perhaps, if IBM, Apple, and Fuji Xerox got together again (Unicode), they could actually make universal speech recognition a success.

[macFanDave]

I wish I could get Voice Commands to work reliably on my wife. When they do work, things REALLY slow down.


Excellent! When I am trying to watch TV and my wife starts jibber-jabbering I keep pointing the remote at her and hitting the Mute key furiously. I'll report back if that EVER works!

Also, I started using Voice Recognition with a Performa 6400 (Single 0.200 GHz G2 processor, 0.084 GB RAM), OS 7.5 and 8 back in 1997. I find it very depressing that its ability to understand me hasn't changed at all apparently on a new PowerMac (2 x 2GHz G5, 1.5 GB RAM) running Tiger. The text-to-speech is not much better either.

To me it seems the whole speech thing (recognition AND synthesis) is a technology Apple has abandoned and kept around for the few eccentrics that use it. On the other hand, these technologies haven't gained much traction in the broader industry.