Speech Recognition

1
Speech Recognition

• Speech Recognition is the use of computers to
  hear and understand spoken words. 
• Although some desktop software may incorporate
  speech recognition, it has not been as
  beneficial for routine desktop applications.

2
Speech Recognition

• Speech is the bicycle of user-interface design.
  It is great fun to use and has an important role,
  but it can carry only a light load. Sober
  advocates know that it will be tough to replace
  the automobile graphic user interfaces.
  (Shneiderman)
• We are not yet to the point of the HAL 9000

3
Speech Recognition

• Why is it so difficult? Speaking commands is
  more demanding of working memory than is the hand
  eye coordination necessary for mouse pointing.
• For designers of human-computer interaction,
  speech technology has four variations
  discrete-word recognition, continuous speech
  recognition, speech store and forward, and speech
  generation. (Shneiderman)

4
Discrete word recognition individual words
spoken by an individual person. These have
90-98 reliability, most require speaker
training, though some commercial products soon
may be speaker-independent. Examples include
aircraft engine inspectors, baggage handlers,
wireless VCR controllers, phone voice dialing
services. Some medical report generators will
recognize key words then generate a boilerplate
or template document. Speech-enabled automobile
phones and accessories are already available. 
Toys, games, and both industrial and household
automation commanded by voice are in the offing.
5
Continuous speech recognition The commercially
successful products are still limited to
specialty niches, such as radiologists. The
difficulty is that normal speech patterns blur
the boundaries between words. Again, these have
a 90-98 reliability and nearly all products
require speaker training.
6
Speech store and forward ie, VoiceMail. This
technology is reliable, low cost, and well liked
by users. Speech generation devices speech
synthesis devices are used inexpensively in
automobiles, cameras, games, vending machines,
and cockpits. The Kurzweil Reader is an
application for the blind, to convert text into
speech. Synthetic speech can be annoying, and
therefore, digitized human speech is sometimes
preferred. Speech segments can then be
concatenated to form more complex phrases or
sentences.
7
Does it really work?      

• "Speech recognition has gone from the bleeding
  edge to the leading edge . . . The top 25 percent
  of large companies are implementing their
  first-level speech applications because in the
  near term, speech recognition provides a bigger
  gain than even the Web. That's how customers
  contact and interact with businesses." 
• Brian Bischoff, ATT, quoted in  Information
  Week, February 22, 1999.

8
Speech Recognition and the Internet?

• Phone Access To The Web
• Talking and Listening to the Web From Your
  Desktop
• Dictation and Transcription
• Call Centers and Telephony Applications
• Accomodation For Disabled Persons

9
Dragon NaturallySpeaking

• supports virtually all Windows applications.
  Initial training can be completed in only five
  minutes. Users age 9 and older can speak at a
  normal pace - up to 160 words per minute. Your
  speech is transcribed immediately, appearing as
  text on the screen and in reports, letters,
  e-mail messages, chat rooms, and Instant
  Messaging windows. Format and edit documents by
  voice. Navigate the Internet by speaking URLs and
  into fields in Web pages. Navigate and control
  the desktop by speaking drop-down menu commands.

10
Dragon NaturallySpeaking

• HIGH ACCURACY
• BROAD COMPATIBILITY
• FAST SETUP
• INTERNET-FRIENDLY
• VERY LARGE VOCABULARY
• INTEGRATED WITH WORD
• NATURAL LANGUAGE COMMANDS
• CONTROL THE DESKTOP
• SELECT-AND-SAY EDITING
• MULTIPLE USER SUPPORT

11
Dragon Medical Suite

• Medication names
• Medical procedures
• Medical diagnoses
• Diseases
• Dictating on the Go

12
Examples of commands

• To make a table
• Create a table with four rows and five columns.
• Add a four by five table.
• Insert table five columns by four rows
• To change a color
• Turn the next paragraph red.
• Color the following paragraph red.
• Make next paragraph red.
• Select the following paragraph. Make it red.

13
Example of Macro

• Request Lab Tests
• Dragon NaturallySpeaking Medical Suite opens up a
  standard request form for laboratory services.
  You say "Bill Smith" and Dragon NaturallySpeaking
  Medical Suite fills in the patient information.
  You then request "full workup" and "send results
  to office" to complete request.

14
PC Magazine Online

• Editors Choice
• Downloads