Speech Recognition Suggestions

During December of 2003, the Boston Voice Users group compiled a list of suggestions for improvements to speech recognition software, drawn from members of our group, as well as world-wide speech-users mailing lists. At our December meeting, we discussed the suggestions, and held a vote to rank them. Listed below are our top ten suggestions, together with a supplemental list (also ranked) of additional items. We have also included lists of suggestions related to IBM ViaVoice and to Microsoft's speech software, although there were not enough users of either present at the meeting for it to make sense to rank the suggestions. The New York Speech Recognition Users Group has also created a wish list, combining our list with the added their own suggestions and compiled their own ranking. Each of these lists has been submitted to the appropriate companies.

Top 10 suggestions:

1. An email suggestion box that is separate from tech support (no personal response necessary) This would enable people to send in suggestions without being charged
2. Support for Select and Say email, especially in Eudora
3. A ScanSoft employee who monitors forums and voice groups
4. A ScanSoft employee whose job description includes using NatSpeak all the time
5. Implement "Correct That" and "Scratch That" using the old "backspace" method rather than the version 7 method using special Windows keystrokes. Before version 7, "Correct That" and "Scratch That" worked by sending enough "backspace" characters to delete the previous utterance. Version 7 uses a fancier method using special Windows keystrokes, but this approach fails in applications that don't support the special keystrokes. Let's return to the "backspace" solution for all applications, or at least give us a way to specify that a particular application should use the "backspace" solution.
6. Fix performance problems with large MS Word documents. NatSpeak gets very slow when dealing with large documents and moderate sized tables.
7. Improve recognition for common homonyms like to/too/two/2, for/four/4, and one/won
8. A Global command for scrolling that also allows users to vary the speed of scrolling
9. An "X Replace With Y" command that would combine selecting a word or phrase (x) and replacing it with another word or phrase (y). This would cut in half the number of commands required to edit a document, and would make editing cognitively easier and more comfortable by taking away the waiting periods that normally occur between selecting text and changing it.
10. A set of "Try Again" commands that would combine undoing with making another attempt and at the same increase the possibility that the second attempt would be correct by allowing the user to give the computer a contextual clue. "Try Again" would simply return the next choice in a correction box, "Try Again Text" would return only text-mode choices "Try Again Command" would limit the choices to commands. Further possibilities: Try Again Written, Try Again Numeric, Try Again Homonym, Try Again Symbolic, Try Again a-z... (This would allow you to start to spell a word.)

Supplemental list of suggestions aimed at naturally speaking and ranked by importance by the Boston users group:

• A spoken command to hold a key down (e.g. control) during the following spoken command.
• A way to share all vocabulary modifications between users, including word properties and deleted words
• A way to speak a sequence of commands without pausing.
• An option to force NatSpeak to run entirely in memory and not swap to virtual memory.
• A fix for problems with NatSpeak adding stray "ed" and "s" endings.
• A way to specify whether to favor abbreviation or full spelling that is broken down by types of abbreviation and/or specific word.
• Add "Numeral" commands for two-digit numbers (e.g. "Numeral 10")
• A link that allows you to open a macro file by clicking on a command in the command history dialog box
• A way to list all commands that contain given words or are about a given subject (e.g. List all commands that have to do with selecting lines)
• A correction box option to learn a word without capitalization in a context where the word would have been capitalized (e.g. after "cap" or "period") so that the corrected word is not added to the vocabulary as a capitalized word
• A fix for the problem in Microsoft Word of periodic loss of connection to the text, which disables the Select and Say commands
• A fix for the recognition problem between "and" and "end"
• A fix for the problem of a current window losing focus when there is no reason for it to have lost focus (this must be corrected by clicking the mouse in the window, which only sometimes works, or switching to another program and back)
• A strong correction option in the correction box to learn after 1 correction as if you had corrected 10 times
• A "Nothing" correction option to fix problems with noise misrecognized as words
• A fix for the spacing problem in programs that do not support select and say (e.g. Eudora)
• A way to use typographic quotes and dash (not " and --) in Microsoft Word.
• A fix for the comma misrecognition problem: "," is recognized when "come" or "common" is spoken
• Spoken in-line commands to switch to text-only or command only mode for a following command ("Text XYZ" and "Command XYZ")
• A fix for problems with small words getting lost
• A fix for problems with small words appearing when they shouldn't
• PowerPoint compatibility, at least in the notes window
• An option to favor numerals vs. written numbers
• Commands that make the command browser usable hands-free
• A way to disable/enable NaturalText by application
• Commands to scroll left and right
• Commands to tab or "scratch that" multiple times
• An option to increase the font size in correction dialogs
• An easier way to disable built-in commands or at least change their names
• A way to turn off a single or a set of installed macros
• A "Restart NatSpeak" command
• A fix for the problem with "Enter" sometimes not being recognized when an item is highlighted
• A way to "select next 3 lines" in Windows Explorer
• A way to preserve the formatting of original word when selecting a word, then dictating another word
• A dictionary that is updated, synchronized with MS Word dictionary, and/or a choice of dictionaries
• Bring back the "Go to Favorites " command
• A way to control the priority of common phrases; this would allow a user to tell the program to give less weight to common phrases so there would be less likelihood that the program would interpret a phrase as a common phrase
• Better recognition with some build-in commands (e.g. "select 9" vs. "select line")
• Improved accuracy for female speakers
• An easy way to switch among multiple web browser windows
• Better recognition logic or an option that will cut down on misrecognitions that are ungrammatical ("he walk")
• An option to change the "Wake Up" or "Listen to Me" commands
• An option to provide confirmation for "microphone off" command

• A way to dictate a whole word instead of spelling when in the spell dialogue box
• A way to assign a set of macros to multiple programs

Suggestions for ViaVoice: (not ranked)

• Frequently makes unwanted contractions and never learns
• End of dictated sentence doesn't appear
• Misplaced or unwanted capitalizations mid-sentence
• Indicate that a phrase you never use shouldn't be recognized
• The medical vocabulary should be updated
• Ship with Susan Fulton's "always active" nav macros
• Correction Window won't recognize words/phrases
• Correction Window says "nothing selected" when something is
• Sluggish performance dictating into MS Word
• Performance problems with specific applications
• Eliminate clashes with firewall & anti-virus programs (Norton) UI
• Alphabetize all words in vocabulary manager
• Allow enabling or disabling tray below correction window
• Assign Mic On/Off to a single key
• The command key for Word ought to really work
• Dictation Macro Editor: choose macro to edit by typing a few letters
• Lose the agents (too cute)
• Clarify whether Recognition Sensitivity instructions affect enrollments or command recognition
• Dictation doesn't appear at cursor or replace selection
• Program crashes, losing dictation
• An autosave feature in SpeakPad
• SpeakPad template fields: capitalize first word of sentence if field follows a period or paragraph mark

Suggestions for Microsoft Speech: (not ranked)

• Create dictation and navigation macros
• Documentation
• Fuller manual
• More general information (nature of engine, size of vocabulary, options of augmenting vocabulary)
• Fully reliable list of available commands
• Support for UK English
• Support for major European languages
• Support for Swedish and Catalan
• One single program for all languages
• Dictate in several languages in the same document
• Transcription of voice dictated files
• Save voice files for transfer/restoration
• Export and import custom words and customized commands

Return to Boston Voice Users home page

This document last modified on 1/5/2004

Problems with this page? Contact web@bostonvoiceusers.com