- Computers & Software»
- Computer Software
Customer Review of Nuance Dragon Naturally Speaking Voice Recognition Software
Review of Version 13: Very Pleased
This is what I wrote to my siblings soon after I bought Nuance Dragon Naturally Speaking Version 13 Premium Edition in March of 2015: "My new Dragon 13 on my new desktop computer is wonderful. Versions 9, 10, 11, and 12 barely competed with my slow, error-prone typing. The error rate decreased over the years with practice, training, and program improvements but were still rather high in version 12. My main gripe was the long 'thinking time' between my dictating and Dragon 12 typing. I installed version 13 in my new desktop computer fresh rather than as an update and without my old profile file. Even with no training, aside from having me read aloud a few paragraphs, it makes far fewer errors than version 12 did even after years of training it. Best of all, version 13 responds quickly, with hardly any, if any, 'thinking time' between my speaking and it typing. And Dragon now works directly in my most used programs—Open Office Writer and Gmail—rather than via the Dictation Box. I have been zipping through my revision edits of a short story in progress. I conquered a long list of emails needing replies and have been replying to emails within a day or two. I anticipate being very productive as a writer compared to when using my laptop computer and Dragon 12.
"Why is Dragon 13 the most accurate and fastest version yet? One factor is that correcting errors while using Dragon 12 has taught me to speak more distinctly, because when making a correction I listen to a playback of how I said a word and can tell if the mistake was my fault because I slurred my speech. And much of the credit for the increase in speed and accuracy doubtless goes to the program. Each version is much better than the previous version because the company has more speech samples for its software to analyze word use probabilities. But the main factor, I'm sure, is having a very fast computer processor and ample RAM."
Earlier that month, when my laptop computer was giving me trouble, with family help I bought an on-sale ASUS M32AD-US003S desktop computer system with a 3.5 GHz Intel Core i3-4150 Processor, 8GB DDR3, 2TB HDD, running Windows 8, which I later updated to Windows 10. I bought a monitor and whatever else I needed to complete the system. Having made sure to buy a computer system with more than enough power and speed for it, I bought version 13 of Dragon NaturallySpeaking for half price from an online retailer.
Update of January 2016
The current version of Dragon NaturallySpeaking is 13. I have been very pleased with it, so I don't expect to upgrade to version 14 when available. It's wonderful to dictate paragraphs with few errors.
A significant new development in speech recognition software is the coming of Google speech recognition. It can be used as an add-on to dictate text in Google Docs. I have of late often instead been using the Chrome browser app Speechnotes, which uses Google's speech-recognition engines. Of course, Dragon NaturallySpeaking is far better. You can, for instance, customize it to your voice and vocabulary. But the Google Docs speech recognition add-on and the Chrome app Speechwriter are free, so I forgive their making a lot more errors. They are accurate enough to be faster than my so-so typing speed. That is not true of the speech recognition program included with current versions of Microsoft Windows, which for me performs so poorly that it's useless. Whenever my circumstances allow, I use Dragon NaturallySpeaking Premium Edition 13 on my desktop computer system that has ample memory and speed, and when I have to use my notebook computer or my father-in-law's old desktop system, I use Speechnotes.
Either way, I have an agreement with my wife to dictate-type only when she is not home or is in another room with her hearing aids out or with the door closed.
Customizing the Dragon Vocabulary
All of the versions and editions of Dragon NaturallySpeaking that I have used have had the most difficulty gtetting small, common words like conjunctions and prepositions right. A big reason that I have been having better success with the software than in years past is that I gave up trying over and over to teach Dragon to tell 'and' from 'in' and so on. Also problematic are such sound alike words as to, too, and two and for, four, and fore. The solution is to teach Dragon to type the desired word when I dictate an entirely different word. Sometimes I make up a word, but more often I delete a distinctive proper name from the vocabulary editor and put it back in as the sound for typing a word the program often gets wrong when dictated normally. This is an ongoing process of trial and error as I search for substitutions that the software consistently types correctly. The table below shows my latest set of substitute words.
a = Aristotle
in = innyzinny
there = Tchaikovksky
an = Andover
kept = zeptizoopy
their = Tulsa
and = Gretel
on = ondondoe
to = Toledo
are = arfabarf
or = ornogorno
too = Thelonius
for = Desdemona
OR = Orlando
two = Tupelo
About My Experiences with Earlier Editions
Below are paragraphs I posted years ago about earlier editions of Nuance Dragon NaturallySpeaking. I am keeping them as an archival record of my long and difficult pursuit of definitely better than typing speech recognition software. What finally brought me to that goal? The improvements in the program from version 9 to version 12 and beyond were major factors and so was my training and customizing of the software to suit me and so was my buying a more powerful, faster computer (for a bargain price) and a good quality for the price headset and so has been my improving my speech habits. Often the software I did not understand me because, unawares, I slurred and clipped my speech. Listening to myself whenever I correct an error has little by little improved my diction. Now that I am a much faster typist using speech recognition software, I have no excuse for not being more productive as the writer.
My User Review of Dragon Naturally Speaking 12, updated near the end of 2013
I bought Nuance Dragon Naturally Speaking Version 12 Premium Edition in August 2012, on sale then at a tempting price for past customers, after reading positive reviews about its superiority to Version 11. My old laptop computer did not meet version 12's minimum requirements for processing speed, so I installed it on my wife's laptop computer. She was not interested in trying speech to text software, and it turned out not to be convenient for me to fit in time on her computer when she was not using it or to learn the quirks of her computer, so I continued to use version 11 and rarely used version 12. Then, in October 2013 a relation gave me a much faster laptop computer, running at 2.1 GHz with 4GB RAM, compared with 1.5 GHz with 4GB RAM on my wife's computer and 1.73 GHz with 1.99GB RAM on my old computer. I uninstalled Nuance Dragon NaturallySpeaking Premium Edition version 12 from my wife's computer and installed it on my newly acquired computer.
Once I was running it on a suitably fast computer of my own, I found Premium Edition version 12 to be significantly better than Home Edition version 11, which I already liked. From the start version 12, with just a little training of a new voice profile, converted my speech to text more accurately, with fewer errors, than version 11 did after many months of training and corrections. And the more I corrected errors, the better it got. That was also true, but I subjectively think much less so, of 11.
While far fewer, version 12 also made quite a few mistakes. As with earlier versions, especially troublesome were common short words. The program just could not consistently distinguish between and / an / in, is / his, as / has, at / that, and so on. To get around that problem, I added made up words to the Vocabulary Editor that were distinctive in sound but had the same spelling as common words. I've shown some examples in the table below.
The fault is more mine than the program's. I have never mastered speaking clearly. To me when I speak, my words are very clear, but when I ask the program to play what I said back to me and I listen to myself as it heard me, I realize that I did not speak at all clearly, that I chopped off the ends of words and slurred word together. I conclude that most people will get even better results than I do, because they talk better. My enunciation is improving (such as distinctly pronouncing the "ed" at the end of past tense verbs or saying "in a" instead of grunt), so the program is making fewer errors.
I created two other substitutions -- to print the word 'kept' I say zeptizoopy and to print the word 'period' I say pubblup -- because the program kept mistaking the word 'kept' for the command cap, which tells it to capitalize the next word, and kept as usual printing a punctuation mark period when I said period even when I wanted it to print the word period. The program is supposed to know that you are giving a command if you hold down the CTRL key and to know that you are dictating if you hold down the SHIFT key, but neither trick works for me when I say period.
To distinguish between the plural and the possessive of a word, I pronounce the ending s if it is plural (such as "cats") and if it ends in 's, I say the word and then say "apostrophe s" (like, to print "cat's" I say "cat apostrophe s"). I changed the Properties for the apostrophe ' in the Vocabulary Editor so that no space comes before it.
Evidence that buying, learning, and training Dragon NaturallySpeaking Premium Edition version 12 is worth the cost and bother is that I have tried several times in the past to do NaNoWriMo (National Novel Writing Month), the object of which is to write the first draft of the first 50,000 words of a novel in the 30 days of November. I have never before come close to achieving the goal. One year I wrote only 5,000 words in 30 days, only one tenth of the goal. Another year I reached 25,000 words, only halfway to the goal. Other years I have been so far behind from the first day that I soon gave up. This year, 2013, between November 1 and November 30 I "typed" (by dictating to Dragon Naturally Speaking 12) over 50,200 rough draft words. For a change, I wrote short essays instead of a novel. Most days all month I was way ahead of schedule, and when obligations kept me away from my computer and my word count fell behind, I quickly caught up.
The cheap headset microphone that came with version 11 when I bought it long ago soon afterward stopped working. Thereafter I used for a microphone my old Logitech webcam that I found in a drawer in my desk. It seemed to work tolerably well for dictating. I had excuses for putting off buying a microphone recommended by Nuance. Ours is a rather low income household, and there were always higher spending priorities, such as paying the rent and feeding the cats. I also worried that a microphone good enough to make a big difference would cost hundreds of dollars. But the main reason was that, as an enneagram 9 personality type, I tend to procrastinate taking care of business. Finally in November 2013 I got around to looking at the page on the Nuance website that lists the microphones that they have tested and found to work excellently with their software. I discovered that one of these, the Logitech USB Headset H340, retails for $29.99 and was on sale at a local store for $19.99. It showed the same test score as headsets costing over $100. So I bought one and am very happy with it. My dictation to text seems many times more accurate using it than when I was using a Logitech webcam microphone. Getting a better microphone may have brought an even bigger improvement in my dictation results than did upgrading to Premium Edition version 12.
According to one review I saw, the headset that comes included with some version 12 deals is of excellent quality. Of course research further if that is so before you buy.
Here is a review comparing Dragon Naturally Speaking with other voice recognition software.
Version 11.5 Introduction
I'm loving my Nuance Dragon NaturallySpeaking Home Edition Version 11.5 speech-recognition, voice-to-text software. I speak into the microphone of the headset that came with the program, and the program types what I say. It does not actually move the keys on the keyboard of my computer as though with invisible fingers, but the words I say appear as text on the monitor screen just after I speak, as though I were typing them instead of speaking them.
Windows speech recognition versus Dragon Naturally Speaking
My Experience with Dragon NaturallySpeaking 11.5
The following paragraph is just as Dragon NaturallySpeaking 11.5 transcribed it when I dictated it one day in the third week of February 2012, and beneath that is the edited, corrected copy of the same paragraph.
Dragon NaturallySpeaking home edition version 11.5 worked at least this will let us a box has version 9 did after by nine months of use. Even better. It almost always types Kayle when I say Kayle. Getting whole phrases and even whole sentences right hand was much more frequently. But after more than a months of daily use, and then choose it has improved a lot. It still confuses small, common words such as 'and' and 'in'. I struck to the program tonight 'and' when I say 'glubglub' had to write 'and' when I say 'insidious', but it's very hard to remember when talking to make no substitutions as a speak.
Dragon NaturallySpeaking Home Edition version 11.5 worked at least as well right out of the box as version 9 did after about nine months of use. Even better. It almost always types Kayle when I say Kayle. Getting whole phrases and even whole sentences right happens much more frequently. But after more than a month of daily use, I am not sure that it has improved a lot. It still confuses small, common words such as 'and' and 'in'. I instructed the program to write 'and' when I say 'glubglub' and to write 'in' when I say 'insidious', but it's very hard to remember when talking to make those substitutions as I speak.
Compare the before and after paragraphs above. That is how well Dragon NaturallySpeaking 11.5 is working for me after a little over a month of use. Stay tuned for an update every few months of the program's progress in accuracy.
Here are some Dragon NaturallySpeaking 11.5 errors that I have collected. What the program typed is on the left of the equal sign and what I actually said is on the right of the equal sign:
kisses so I would typing them = just as though I were typing them;
Packet 2008 = Back in 2008;
James the. Smith = James V. Smith;
and 2000 = in 2000;
the chapter which Smith = has a chapter in which Smith;
I hell = I held;
stories, essays, and females = stories, essays, and e-mails;
into clubs = into hubs;
when I light longhand = when I write longhand;
inappropriate = and appropriate;
and less = unless;
handle = hand wrote;
community is six justice and peace = community that seeks justice and peace;
and yet they speak of affable fangs = and yet they speak of ineffable things.
The essay "Rest in Peas" by Robert Fortner explains why speech recognition software inevitably makes mistakes, printing to the monitor screen words very different from what the speaker said. When something said might be either, or any, of two or more sound-alike words or phrases, the software uses statistical analysis of speech data (gathered from many speakers and from that individual speaker) to make a bet on which word or phrase was said. The program gets more accurate as it accumulates more data in an individual speaker's speech profile. Over time it will have more data about my speaking habits and its bets about what I said more often will be winners. Better input, such as using a better microphone or enunciating more clearly, would also improve the accuracy of the program. But there will always be errors, because, as poker teaches, what is the statistically best bet is not always what is actual. When the choice for the software is between two common and somewhat similarly sounding words, such as 'in' or 'and', as slurred by me, it's pretty much a tossup which it will guess.(Back to Fortner's essay, here is a response.)
Progress Report on My Dragon Naturally Speaking 11.5 Experience, Added in July 2012
So how well is my Dragon NaturallySpeaking Home Edition performing for me after several months of fairly frequent use? Well enough that I prefer it to typing and look forward each day to chances to use it; poorly enough that I often get irritated with it. It is learning to understand what I say better and better.
* The program apparently cannot learn to consistently distinguish among little, common words, such as an/and, in/and, and so on. My workaround is to instruct the program's Vocabulary Editor that if I say a certain made up word, to type a certain common word. For instance, if I say "akeybrakey", then the program types "a".
* My wife does not like me to be dictating when she is present, so my dictating time is limited to when she is at work, is on errands, or is in another room with her hearing aids out. When she is in our living room (which is where I write and she does art), I type the old-fashioned way.
* I made the mistake of using the Learn from Specific Documents command without thinking through which documents. With that command, Dragon NaturallySpeaking studies the specified documents and adds any unfamiliar words to its vocabulary. I made the foolish mistake of telling it to examine my entire My Documents directory. As a result, thousands of misspelled, run-together, jargon, and foreign words were added to the program's vocabulary, greatly multiplying the odds of its guessing wrong what I say. After weeks of deleting such added custom words from the vocabulary when they caused problems, I finally just deleted the whole custom words vocabulary and started creating an actually useful custom words list from scratch.
Remembering Version 9
I first learned about Dragon NaturallySpeaking back in 2008. At the time I was living in Moscow, Idaho. My wife Kayle Rice was the minister at Unitarian Universalist Church of the Palouse. and I was working on my novel The Son Who Paid Attention (published in 2010). I like to read books on how to write, and I read the book Fiction Writer's Brainstormer by James V. Smith, published by Writer's Digest Books in 2000. The author devotes a chapter to expressing his enthusiasm for voice-recognition, speech-to-text software. If I am remembering right, he says that that is how he does his writing and that dictating to Dragon NaturallySpeaking he can "type" as fast as, if not faster than, the fastest professional typists.
I was very interested in that claim and was immediately sold on the idea of using voice-recognition software. Back in the mid-1970s I taught myself to touch-type, and over the decades I have done a lot of typing, both in my work as an antiquarian bookseller until I retired and as a writer, but I have never been able to type fast. While the top 10% of professional typists can type 65 words per minute or better and the average typist can type in the vicinity of 40 words per minute [Source: "Typing Speed: How Fast is Average: 4,000 typing scores statistically analyzed and interpreted" by Teresia R. Ostrach], my fastest copy typing speed is under 35 words per minute and I can type my thoughts at a little under 20 words per minute. I hoped and expected that Dragon NaturallySpeaking 9 would greatly increase my writing productivity.
I bought and installed it and completed the instructions for training it. The program was a disappointment. The software could not understand what I said. I did not blame the program. My voice has always had its own private accent. I never learned to pronounce my Rs correctly, and from when I was a preschooler, people have complained that I talk too softly or too loudly and that I talk in a monotone and mumble my words. Because I sound perfectly clear to myself, I have never been able to learn to speak like normal people do. When I started using Dragon NaturallySpeaking 9, I soon realized that getting it to transcribe my speech accurately would be as much a matter of training myself as of training the program.
I stubbornly persisted, and after three quarters of a year of nearly daily struggle with the program, it was able very often to turn whole sentences of my dictation into accurate text. The errors that it continued to make were frustrating. It could not get 'Kayle', so I taught it to type 'Kayle' whenever I said 'honeybunch'. It often confused small, common words, for instance mistaking 'in' for 'and' and 'and' for 'in'.
The biggest problem was that my computer system was not powerful enough. The processor speed was barely up to the minimum requirement and was far below the recommended speed. Every time I dictated anything, the program had to think for several seconds. That meant slow typing by dictation even when the program was accurately transcribing my speech.
That was the situation when my desktop computer broke down. The system was too old to be worth repairing. My wife let me use an old laptop of hers. It was not up to the minimum requirements of installing and running Dragon NaturallySpeaking.
Then in the fall or late summer of 2010 I read reviews of Dragon NaturallySpeaking 11. The reviews said that it was far more accurate right out of the box than previous versions had been. I longed to have a computer fast enough that I could buy and try it.
My wish came true in January 2012 when a dear, generous relation shipped me the gift of his Compaq Presario V6000 laptop computer running Windows XP at 1.73 GHz. That is pretty close to the 1.8 GHz recommended speed and much faster than the 1 GHz minimum speed required to run Dragon Naturally Speaking. I scraped together the money to buy Version 11.5 Home Edition on sale. It soon arrived, together with a headset, and I installed it and did the initial training readings.