I suffer from RSI. When it gets really bad I switch to a speech recognition software such as Dragon Naturally Speaking or DragonDictate. Over the years I have tested a lot of speech recognition systems on OS X. To tell the truth I have never been that impressed by any of them. It is the one area in which the Macintosh has lagged behind Windows in terms of software for writers. I started with MacSpeech’s iListen back in the days of the PowerPC iBook. A number of people reported getting good results with it but I’m afraid for me it was a pretty terrible program. It could never do anything I wanted it to do.
This was certainly not true of its Windows counterpart Dragon NaturallySpeaking 8 at that time. Since then MacSpeech has been bought out by Nuance, the makers of Dragon NaturallySpeaking, which has given them access to the same basic speech recognition technology. The first couple of versions of this new program, these days known as DragonDictate, were not terribly impressive either. The basic speech recognition engine was superb – every bit as good as its Windows counterpart but the programs were hampered by a very clunky correction interface (a very important part of a speech recognition program) and a number of really annoying bugs which made the program virtually unusable, at least for me.
So now we have come to the latest iteration of DragonDictate, Version 3. The question is does it improve on its predecessors and can it meet the challenge of its Windows counterpart?
I bought the program is an electronic download and it took several hours to download to my computer. The installation went very smoothly.
I must admit to being initially very impressed by DragonDictate 3. I also own Dragon NaturallySpeaking 12 and performance seems to be equivalent between the two programs. It also has to be said that the correction interface has been much improved and is now very similar to the Windows version – which is no bad thing, believe you me.
Recognition accuracy is superb right out of the box. After 5 minutes of basic training I was getting roughly 99.5% accuracy when dictating in normal business English. For writers there is my one usual caveat – recognition accuracy is based upon the program’s ability to predict what is being said based on word order. This means that the program is more accurate the more commonplace the English that you use. For a writer, looking to find that striking phrase, this is a bad thing because it means the program struggles to recognise what you are saying under those circumstances. Eventually, it will get used to the way you write though, which is why the correction interface is so important—a decent interface really speeds up this vital function. Over time the program will eventually adapt itself to your writing style.
In the past, the Apple version of the software has really struggled to learn the sort of specialised vocabulary such as I use when writing a fantasy novel and I regret to have to say that this is still the case. The Windows version easily learned such new words as Tyrion and Teclis – in case you’re wondering how I managed to get the program to understand those last two names, I used the correction interface. There are some easy workarounds for this, such as using modern names and place names like Terry for Tyrion, London for Ulthuan and so on and then doing a find and replace but it is frustrating that the OSX version remains behind the Windows version on this.
So far, only one of the annoying bugs of the previous version has surfaced. Sometimes when you make a correction, the program will capitalise the first word of the corrected text regardless of whether there should be a capital there or not. I’m actually very pleased with this given the number of truly strange bugs that there were in the previous versions and the alarming regularity with which they arose.
So far, the program seems to be very stable and very accurate and I would not hesitate to recommend using it for dictating an email or a straight business letter. In case you’re wondering, I did dictate this review using the software and so far I have had to make roughly 10 corrections over about 700 words which is recognition accuracy of roughly 98.5% according to my back of an envelope math.
Also in this version is the ability to transcribe speech files, a thing that was only available as a separate program previously although it was incorporated in the PC version.
It also has to be said that there is one area in which the Mac version of the speech recognition software has a distinct advantage over the Windows version. It is possible to program new commands directly into DragonDictate 3 so that you can automate complex tasks and trigger them by voice command. This is something that is only available in Dragon NaturallySpeaking Professional which cost several hundred pounds more than the basic version of the program. I have tried this and it does work.
So, credit where credit is due, this is the first version of DragonDictate I have been happy with. It has taken several iterations of the software and several hundred pounds of my money but, at last, it seems like this and speech recognition has arrived on the Mac. Of course, I am holding off final judgement for a few weeks yet because I have been disappointed by Nuance’s offerings on the Mac before. I may well be again.
Stop Press: After completing this post, I sat down to play with DragonDictate 3 and all of the previous bugs reasserted themselves. I don’t know why this should be. I have tried DragonDictate on three separate versions of OSX and many different Macs. I have done time machine backups, new profiles and even clean installs on new machines and somehow these problems always recur. It may be something to do with my setup, a conflict with some other piece of software I use, but I get lots of strange problems. Sometimes letters are dropped from words, sometimes when I make a correction both the correction and the previous version are left in the text. It is irritating and it undermines my faith in the software. As of this moment, I am, once again, going to have to reluctantly conclude that I would be better off using Dragon Naturally Speaking either on Boot Camp or on a virtual machine. Bummer!
Stop Press 2 (November 23 2012): I believe I have found the source of my problems. I have disabled TextExpander and none of the show-stopper bugs I was experiencing have returned. I have tested this over the past few weeks and dictated more than 10000 words without problems. I have also trained Dragon Dictate 3 using the text files for this blog and it has now learned my specialised Warhammer fantasy vocabulary just as well as Dragon NaturallySpeaking. Under these circumstances I can unreservedly recommend this program.