Installation

Requirements

• Windows XP or Windows 2000
• NaturallySpeaking 7 Preferred (or Professional) edition (or perhaps lower)
• GNU Emacs > 21.3 (or perhaps 21.3).
• AUCLaTeX for LaTeX support (only if necessary)

Install software

• Unpack the distribution into a directory (folder), say c:\Emacs. This will create a subdirectory c:\Emacs\listen.
• In your .emacs file, add the following:

  	(when (eq window-system 'w32)
  
            ;; ADJUST THIS PATH IF NECESSARY --->
  	  (setq listen-path "c:/Emacs/listen")
  	  (setq-default load-path (cons listen-path load-path))
  
  	  ;; EmacsListen will turn it on anyway; best to get used to it
    	  (transient-mark-mode 1)
  
  	  ;; LOAD VR-MODE
  
  	  (load (concat listen-path "/vr-mode/vr"))
  
  	  ;; These commands below may not be desired if you are
  	  ;; already using vr-mode's voice commands
  
  	  (setq vr-activate-minibuffer nil)
  	  (setq vr-activation-list nil)
  	  (setq vr-voice-command-list nil)
  	  (setq vr-command (concat listen-path "/vr-mode/vr.exe"))
  
  	  ;;; LOAD EMACSLISTEN
  
  	  (require 'listen-mode)
  
  	  ;; Load speech engine specific interface, here Barry Jaspan's vr-mode
  	  ;; for ScanSoft NaturallySpeaking as modified to work with EmacsListen
  	  (require 'listen-vr-additions)
  
  	  ;; (Optional) Load listen-peek (display of yank, search histories,
  	  ;; and color coded excerpts of recent views of files)
  	  (require 'listen-peek)
  
  	  ;; (Optional) Activate listen-peek mode
  	  (listen-peek-mode 1)
  
  	  ;; (Optional) Load key and mouse definitions
  	  (load "listen-keys")
  
  	  ;; Recommended (make garbage collection occur less often)
  	  (setq gc-cons-threshold 10000000)
  	)
        

• For e-mail support, you must change the appropriate bindings in the grammar.elg file (for example, replace 'my-read-all-e-mail with 'rmail). And, issue M-x listen-load-grammar to compile and load your changes.
• For LaTeX letterwriting, you might want to use the tech write letter command, which initializes a buffer with your personal information. Customize the variable listen-tex-letter-template for this purpose.
• Note that everything you say (while EmacsListen is listening) ends up in the file named listen-transcript-file (the default is transcript.xml in your listen-path directory). Set this variable to nil if this potential privacy-intrusive behavior is not desired.

Setting up NaturallySpeaking

• Backup your user file or make a copy of the folder (presumably while NaturallySpeaking is not running; this folder is perhaps to be found in C:\Documents and Settings/All Users/Application Data/ScanSoft/NaturallySpeaking/Users).
• Now, NaturallySpeaking must learn about the ShortTalk words that are not part of English. In NaturallySpeaking, choose Words --> Import, and select the file: c:\Emacs\listen\ShortTalk-wordlist.txt
• These words should now be trained. Use Words --> View/Edit and in the Display field, choose "Custom words only". Unless you already have a lot of custom words, just select all of them. Then click on "Train" to provide your pronunciation for each of them. In case you do have a lot of custom words, you may have to chase down individual words manually.
• We can now import the definition of the word sequences that form ShortTalk commands. Apparently, NaturallySpeaking with infer the pronunciation of the sequences now that we have provided a pronunciation of each ShortTalk constituent. So, in NaturallySpeaking, choose Words --> Import, and select the file: c:\Emacs\listen\grammar.txt. This file is very big and it does not contain all possible ShortTalk commands.
• (Optional) It might be useful to build language models that reflect ShortTalk usage. I am not sure about what is the best approach. One idea is to take the transcript files that ShortTalk produces in order to gather a personalized corpus. The XSLT program trainingset.xslt in transcript-programs might be useful. It is also possible that the adaptive algorithms of NaturallySpeaking are sufficient for satisfactory performance. In fact, it may even be possible that the vocabulary building process may hamper performance.
• That's all. Now power up Emacs and issue M-x listen-mode. That will start vr-mode so that all output from NaturallySpeaking is directed to the *EmacsListen* buffer, where EmacsListen parses the text into dictation and commands.