iListen 1.5

It’s a happy day when Mac users have two full-featured speech-recognition programs to choose from–programs that let you dictate using natural speech and control your Mac by voice. This technology is critical to people who have limited use of their arms; for everyone else, it offers an exciting new way to interact with a computer.

Before late 2001, advanced speech recognition was available only in Windows. That’s when IBM released the capable but incomplete ViaVoice for Mac OS X (mmh; Reviews, April 2002). Now MacSpeech’s iListen 1.5.2 joins the Mac OS X speech-recognition ranks. Unfortunately, this program’s performance is inconsistent, but its new ability to transcribe audio files may make it worthwhile to some.

Basic Training

Most speech-recognition programs require a training period during which you acclimate the program to the unique qualities of your voice. With iListen, this process is easy. To begin, you set up the microphone, which in our case was the VXI Parrott with Griffin iMic USB adapter (both included).

Next, for about ten minutes, iListen leads you through sound-level tests and a sample passage of text. With initial training, iListen was about 78 percent accurate in our dictation tests in OS X and OS X’s Classic mode. To improve recognition, you can read 11 other passages. Doing so brought the accuracy up slightly, to about 82 percent. The program can also analyze a set of your documents to learn the vocabulary you use.

Dictation and Correction

iListen requires either OS 9 or OS X, and once you’re finished setting up the program, you can begin dictating into any application, such as Microsoft Word X or 2001. Using an 800MHz iMac G4 with 512MB of RAM, we saw good performance from the program–there was only a slight delay between speaking the words and seeing them appear on screen.

Recognition mistakes are inevitable–and often amusing. After all, English has numerous homonyms (such as flee and flea), and some things just sound alike (for example, workmanlike craft and work them like crap). iListen can correct mistakes–and, most important, learn from them–no matter what application you’re in. Say “Correct that” right after a phrase has been incorrectly recognized, and a correction window opens with a list of fixes. Notably, you can’t make a selection from the list by voice.

And you’ll often spend time correcting the changes implemented during the correction process. For instance, if you stop to delete an extra space or revise an earlier passage while dictating, iListen may lose track of where you are. If so, the program will begin typing in the wrong place. (The PDF manual offers some workarounds for this problem but doesn’t provide a solution.)

Command and Control

Part of speech recognition’s promise is that it can give you the ability to stay off of your keyboard as much as you may need to. Unfortunately, however, iListen reveals some of its roughest edges when you try to control your Mac by voice.

Certain built-in commands work just as you’d expect. You can open a program by saying, for example, “Open Microsoft Word.” However, other commands didn’t work or worked inconsistently in our testing. For instance, “Click mouse” produced no results. Likewise, commands for opening Web pages–“Jump to Macworld,” for example, didn’t work and sometimes caused unexpected results, such as opening the Print dialog box.

iListen relies on AppleScript to do its stuff, which means that if you’re handy with AppleScript, you can write your own scripts. If you’re not, you won’t find many ways to work around iListen’s command-and-control deficiencies. Unlike most Windows-compatible speech-recognition programs, iListen doesn’t give you built-in access to every menu and every key on the keyboard.

The program includes only a few scripted commands. For example, you can say “File Quit” to quit any program and “Return key,” to hit the return key, but you can’t say “Command D” to access Word’s keyboard shortcut for font options. IBM’s ViaVoice doesn’t build in complete keyboard access, either, but it does include an easy-to-use utility that helps you add it yourself. You’ll find a similar (but unsupported) utility called TypeKey Helper in iListen’s Extras folder, but it requires more steps and may be daunting to AppleScript newbies. You must also know that it’s there or stumble across it in the Extras folder–it’s not mentioned in the manual.

The Freedom of Transcription

Despite its many drawbacks, iListen 1.5.2 has one outstanding new feature: it’s the only Mac speech-recognition software able to transcribe AIFF and WAV audio files. This feature alone may appeal to doctors, novelists, and others who prefer to do their writing far from the keyboard. Simply dictate into your digital tape recorder, download the file, push a button, and watch while iListen types what you said.

In our tests, iListen was about 77 percent accurate when transcribing audio files from the iListen-supported Olympus DS-330 digital voice recorder (mmmh; Reviews, October 2002) into Microsoft Word X. However, the program doesn’t offer much help to Mac users keen to improve that number. Unlike Dragon NaturallySpeaking for Windows, which offers the same audio-file-transcription capability, iListen provides no special training mode that will help the program adapt to your tape recorder’s sound quality. You’ll have to make numerous corrections, but many people may prefer that to typing documents from scratch.

Macworld’s Buying Advice

MacSpeech’s iListen 1.5.2 is a work in progress. At this point, IBM’s ViaVoice–although it also has its flaws–is a better choice for most. The primary benefit of iListen today is the ability to transcribe documents dictated into a tape recorder.