MacSpeech Scribe, styling itself as ‘personal transcription for Mac OS X’, is a close relative of MacSpeech Dictate. Both programs use the speech analysis engine from Dragon NaturallySpeaking and both work from samples or profiles of individual voices.
A handy way to describe the difference between the two is to look at Dictate as TV and Scribe as TiVo. Dictate operates in real time with the voice profile’s originator whereas Scribe can be used by anybody, any time.
In Scribe you can create profiles of up to six voices. Whereas Dictate asks you to read some prescribed text into your Mac which it then analyses to create the profile, Scribe asks for an audio file which it processes and corrects with your guidance.
You’ll need a quality audio file of around five minutes for each profile. If you’re creating these files from scratch use a good microphone and reduce ambient noise to a minimum. The file must be in .wav, .aif, .aiff, .m4v, .mp4, or .m4a formats.
You can easily create a suitable audio file with Garage Band or Audacity on your Mac with a USB noise-cancelling microphone (recommended). Or on your iPhone or iPod touch (with an external mic).
As the file is processed the results appear as text in the Transcription Training window. Clicking and highlighting text brings up the Correction window. Use the Play button to hear the highlighted text and make corrections where necessary. When this is done, click Accept.
Then select the whole file and hit Transcribe. As the text tumbles onto your screen you can use the Recognition window to make any further corrections. When you’re done simply save the file or copy and paste the results into a word processor.
As I already use Mac Speech Dictate I plan to use Scribe for externally recorded memos, so I recorded a five-minute file on a second-gen iPod touch running iOS 3.0, and a Griffin iTalk microphone. After training on a 90-second segment Scribe streamed the rest of the text almost word-perfect onto the screen.
To create a second profile I used a female voiceover segment from a documentary I’m working on. This was a more complex file with some unusual words and needed a few corrections. When I’d done this I transcribed a second audio file and the results were dramatically improved.
You can do more training at any time and there are two other tools to improve accuracy. In Vocabulary Editor you teach the program new words. Like Dictate, Scribe also has a vocabulary training module in which you import documents you have written. It then analyses these and ‘learns’ more about the way you put words together. A little extra training effort certainly pays dividends.
First impressions? Scribe does what it claims and I’ve certainly got plenty of work for it to do.
Scribe is not designed to distinguish between multiple voices so you can’t use it to transcribe an interview, for example.
But if the interview predominantly features one voice it should be possible to create a profile for that voice to transcribe particular sections.
Scribe needs an Intel-based Mac with Mac OS X 10.6, and an internet connection is required for product registration. It costs $199 and is available from www.macspeech.com.au.
This article originally appeared in the July issue of Australian Macworld magazine.