IBM voice input package is OK for nontypists

Nontypists will prefer Simply Speaking to one-finger keyboard entry. Decent typists
probably will find this Microsoft Windows 95 program more trouble than it's worth.


To translate what you say into text, Simply Speaking comes with a microphone headset
and VoicePad, a re-engineered version of Microsoft Windows' WordPad.


Simply Speaking does not let you launch applications verbally or dictate directly into
a word processor like its big brother, VoiceType Dictation. Instead, you must use VoicePad
to carry out all speech-to-text translations.


Getting started with the VoicePad interface is as easy as clicking a button. Once
you're ready to export your text into another program, VoicePad's Transfer option will
copy and paste into the desired application.


The included microphone headset performed much better than two low-tech mikes I tried,
although VoiceType is supposed to work with any microphone.


IBM claims 90 percent speech recognition straight out of the box, but that figure
begins to look suspect when you start using the program. The 20,000-word dictionary does
well with conjunctions, prepositions, nouns and verbs in the present tense. When you use
other tenses, it has trouble identifying the word.


Any slight change in inflection or mike distortion will throw the program off. I
suspect the average user will get only about 70 percent to 75 percent recognition accuracy
at first.


But occasionally, Simply Speaking achieves recognition of close to 90 percent.


IBM provides two training features. The first, called enrollment, requires you to speak
a series of set sentences into a microphone so Simply Speaking can construct a language
model of your speech patterns.


Once the model exists, Simply Speaking is supposed to better understand the nuances of
your voice.


There are 283 sentences in the enrollment set, but Simply Speaking can create a small
language model after 50 have been recorded. You can go back later and read the rest of the
233 sentences to improve the language model. This is a tedious process.


To get the best out of the enrollment, repeat each sentence until Simply Speaking
recognizes every word.


You may have to repeat a sentence as many as 10 times if the program hangs up on
pronunciation of just one word.


The second way Simply Speaking boosts accuracy is through a correction feature. When
you're finished dictating, highlight any words that the application incorrectly recognized
and fix them with the Correct Error option. You can add 60,000 words to Simply Speaking's
20,000-word database.


The Correct Error feature generally worked well, yet even after certain words were
corrected, the program did not recognize them. I repeatedly corrected my name, Zach Lutz,
but it still returned text such as "Zach loops," "lacked loops" and
"that looms."


VoiceType Simply Speaking is a promising tool, but it's not a practical application
yet. Speech-to-text translation is a complicated and difficult task. That the $50 package
functions as well as it does is a testament to its developers.


Zach Lutz is a free-lance writer and World Wide Web page designer in Arlington, Va.


inside gcn

  • IoT security

    A 'seal of approval' for IoT security?

Reader Comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above