	Sign In Sign-Up

My efforts are dedicated to my "Source of Inspiration..."

Wow anim1058.gif (13911 bytes)
[ BASICS ] [ Architecture ] [ Text To Speech ] [ Speech Recognition ] [ Microphones ] [ Test Cases ] [ For OEM's ] [ Future ] [ SAPI4.0 ] [ Wow ]

Designing Web Pages to be Speech Enabled

Speech Enabled Web Browses

Every web browser has companion applications that "speech enable" it, allowing users to speak the name of a link and have the browser jump to that link as though it were clicked. The speech functionality will be automatic, requiring no special effort from the web-page designer for recognition to work satisfactorily. However, although speech recognition will work automatically, a web designer can use a little bit of care and get speech recognition to work even better.

To understand what a web designer should do to make his/her pages work well with speech, it's necessary to describe what speech recognition needs to work, and how the web browses automatically speech-enable a page.

For speech recognition to work, it must know what phrases a user can speak, such as "Go to Microsoft's Web Page" or "Tell me about macaws". A speech recognizer can listen for several hundred phrases at a time, and new phrases can be loaded and unloaded at any time. For a the computer to recognize a user's speech, the user must speak the full phrase exactly as it is written, without having words inserted or deleted. If the user says a phrase not on the list, the recognizer returns an "unrecognized" and the user has to rephrase his/her command. If the user said "Go to Microsoft's Page" then the phrase, "Go to Microsoft's Web Page" wouldn't be properly recognized.

Speech enabled web browsers tell the speech recognizer to listen for all the links on a page. They use the same string as is displayed to the user. If the user speaks the full name then the recognizer will hear it, and the browser will jump to that link.

Fine-tuning a web page for speech

Here's a simple rule to fine-tune a web-page to work with speech-enabled browser:

All of the link names must be "speakable".

A link named "http://www.Microsoft.com" is not very speakable because it has to be spoken as "h tee tee pee double-you double-you double-you dot Microsoft dot com". Try saying that. Also, the user doesn't know if they should speak the "h tee tee pee" part. If the link were simply "Go to Microsoft", the user could just speak that.

Now that you know the general principle, here are some tips and tricks:

Don't have more than 100 speakable items

Speech recognition accuracy and speed will deteriorate if there are more than a hundred (approximately) speakable items on a web page.

Don't give speakable items the same name

If two or more links have the same name then the computer won't be able to tell which link the user meant.

Don't give speakable items similar sounding names

Speech recognition has a hard time hearing the differences between similar sounding phrases. Example, speech recognition will confuse "Send mail to Kim" and "Send mail to Jim" easily. The more different the phrases sound, the better the accuracy will be.

Make speakable items two to four words long

If a speakable item is just one word long then it's likely to sound similar to another item. If it's two or more words long then the phrases will sound different, and recognition accuracy will be higher. However, if a phrase is too long, people won't want to speak it.

Don't break a link into multiple lines

Some web pages will break a link into several lines, like this:

My favorite link

to my favorite

spot

Or, the page will do "cool" capitalization like this:

My Favorite Link

Neither of these produce very speakable results. In the first case the user has to say "My favorite link", "to my favorite", or "spot". In the second case, the user would have to speak "M", "y", "F", "avorite", "L", or "ink" to get a recognition.

Don't put ambiguous symbols and numbers in links

A link named "6/23" presents a problem because users are unsure if they need to speak, "six slash twenty three", "six twenty three", "six twenty thirds", or "June twenty third".

Graphics should be speakable

In the case of graphics, you can tell what the browser expects to hear by:

Run Internet Explorer 3.0 (or higher)
Find the graphic or link in question.
Move the cursor over it and let it sit.
Within a few seconds, a tool-tip will pop up displaying the text. This is what the user needs to speak.

Users will speak the text in the graphic. If the text in the tool-tip doesn't match the computer isn't going to understand the user.

Make the title speakable

All web pages have a title that is used for the "favorites" list by browser applications. The Microsoft Voice application allows user to say "Go to <link>", where <link> is any of the names on the favorites list. If the page has a lengthy title or one with ambiguous symbols, then the user will have a difficult time speaking it.

ActiveX Control Design

ActiveX controls should comply with the ActiveX Accessibility standard. Any ActiveX control that supports this will automatically have speakable items.

Using the Speech APIs through Visual Basic or Java

If you want even more speech recognition (or text to speech) than the browser will automatically provide, then you can use ActiveX in Visual-Basic Script or Java Script.

[ Home ] [ Up ] [ BASICS ] [ Architecture ] [ Text To Speech ] [ Speech Recognition ] [ Microphones ] [ Test Cases ] [ For OEM's ] [ Future ] [ SAPI4.0 ] [ Wow ]