MacSpeech Dictate: some training required

Macspeech Dictate is an application that runs on Mac OS X, allowing you to dictate texts on your Mac. The program is relatively accurate but will nevertheless make mistakes — about 80% to 95% accuracy can be easily achieved. The trick is to speak in a natural tone of voice, to have the right microphone, to articulate well without exaggeration. The software itself comes with a Plantronics .Audio headset, which can be upgraded.

I installed MacSpeech Dictate on a Mac Mini. The installation requires you to install some language dictionaries; it will then start a training to learn your voice and pronunciation. MacSpeech Dictate comes in several versions, including the general speech one that I received. There are also bundles for medical and legal applications. I assume these are more accurate out-of-the-box as the dictionaries included with those versions use a limited number of words and expressions.

To train the software I used both British and US English variations to see how much difference it would make, and I used two different headsets; the delivered .Audio headset and the Savi W430-M I still have from my latest Plantronics review. The Savi W430-M performed the best, adding to my comfort as well because it’s a wireless headset (DECT — read the review here). With both microphones, I passed the training session tests without any problem… and that proved to be a problem all by itself.

The fact of the matter is: the training sessions won’t tell you how the software will actually perform under real-world circumstances, and there can be quite a large gap between Dictate’s performance when you’re reading the assistant’s text out loud — with every word and sentence being correctly “understood” by MacSpeech Dictate — and your actual accuracy score when dictating.

When you are training the software, you are reading, and that’s where the problem is. When I’m writing — and a fortiori when I’m dictating — I tend to pause a lot in-between words. I tend to be sloppy with my pronunciation, and last but certainly not least: I’m not a native English speaker, so my English won’t ever be the same as that of native English speakers.

Apparently, all these factors together will confuse MacSpeech Dictate enough to decrease accuracy considerably after using it for more than just a couple of sentences and without further training. In some cases, I got really frustrated and started to type while still occasionally dictating some words as well. This is a surefire way to confuse the software even further — it says so in the manual which warns against doing just that.

Some additional training required

You can train the software to interpret your own way of speaking better, but if you have the awkward tendency like myself to say words a bit different depending on your mood or the weather, or whatever, there’s little use in doing so. On the more useful side, you can create your own dictionaries, so that MacSpeech Dictate can be made to perform more or less as if it is working in a controlled dictionary environment, generating better results. Last but not least, MacSpeech Dictate has a tool to analyse your writing style — you add documents to a vocabulary training window, and Dictate will analyse your style, words, etc. This too will make the application more accurate.

The target application you use matters too. I had the worst performance in BBEdit with that program’s auto-completion feature set to ON, and the best in Dictate’s own Notes window. In-between I found scores that were quite acceptable when using Microsoft Word, and scores which got worse when using applications that seem to interfere with MacSpeech Dictate’s recognition algorithm.

MacSpeech Dictate comes with an abundant set of commands which allow you to command your Mac software (all of it) by voice. This worked well under all circumstances (probably due to the very limited dictionary set the software needs to understand), but oddly enough sometimes added to the frustration when dictating. For example, in order to delete words in a dictated text you must remember some of these commands — actually, quite a lot of them — in order to get some speed out of it.

If you don’t speak those commands (“Delete the Word X to Y”) fast enough, the software thinks you’re dictating again, and will literally type what you commanded it to do. Unfortunately, if you only remember part of the command sentence, that’s just what is going to happen.

In my tests, I experienced better performance when I talked in a steady pace, rather than dictating from the top of my head. This did require me to prepare my text writing by first jotting down some notes (on paper, with a pen — writing as it’s supposed to be…). MacSpeech Dictate was then able to understand most of what I talked about. Without wanting to be accurate, I think the program can be made to perform with a 98% accuracy. That’s still two wrong words per one hundred, but it sure beats the errors I make when typing.

To me, MacSpeech Dictate can therefore not be a complete replacement for a keyboard, but it can take away some of the frustration of typing text; it can actually even improve on my workflow because it allows me to quickly dictate a piece and then go over it again in a (needed) correction round that allows me to improve on my writing.

How efficient is dictating in an office environment anyway?

The number one criticism one reads all over the web when reading reviews of dictation software is that the reviewer can’t imagine a bunch of people all reading their stuff into one a microphone. Imagine the noise!

From what I’ve experienced I’d say these reviewers haven’t really tried out the software long enough or intensely enough to know what they’re talking about. First of all, in most offices people talk on the phone. When you talk on the phone you’re not shouting — usually you aren’t. A good example of a crowded landscape office is the newspaper editorial room. It’s noisy, yes, but no less a good atmosphere to work and concentrate.

My own experiences with MacSpeech Dictate tell me you should absolutely not have to raise your voice to have the software recognise your words. On the contrary even, there seems to be a perfect sound level (that is fairly low) at which Dictate will recognise your words the best. When I checked that level using the rough estimate of the System Preferences, about one third of the input level ‘pills’ turned blue. You can try out for yourself how loud that is.

My point is that even in business environments, publishing environments, etc, you can efficiently use MacSpeech Dictate. The problems of background noise can easily be solved by replacing the .Audio headset with a more sophisticated headset. That is also why I tried out the software with the Savi W430-M, which is a true business system that has sophisticated background noise canceling capabilities.

Especially when used in noisy surroundings, MacSpeech Dictate will benefit greatly from superior noise canceling functionality. To see if my theory was right, I tested Dictate using the Savi W430-M within 1.5 meters of an open window at a very noisy street. It performed just slightly worse than with the window closed — and I do mean slightly…

To round up this review, with a controlled dictionary like in legal or medical environments, I am pretty sure MacSpeech Dictate can achieve very high accuracy scores. With extra training and the user taking the time to create his own dictionaries, I am certain MacSpeech Dictate will perform equally well. MacSpeech Dictate costs around 160.00 Euros. For best performance, buy a professional microphone or headset.