To test the accuracy, I dictated the same paragraph of the text, have used both programs, while speaking normally and without unnatural pauses. I have also decided to carry out the test for either of the programs language training, because I wanted to see what the two do without preparation and with immediate use. Of course it is so that the accuracy of speech recognition improves on the observed level of me away, if you invest time for a workout, so that the program can learn specific qualities of their own voice. However, this also requires time and effort, what the conveniences of voice recognition minimized a bit.
The selected by me text is the first paragraph of the book 'Deep Sea' by James Nestor (by the way: a great book), which reads as follows:
'I am a guest here, a journalist who reported on a sports event which so far hardly anyone has heard: the World Championship in freediving. I am sitting at a small desk in a hotel room on the seafront overlooking the beach promenade in the Greek resort of Kalamata. The hotel is old, you can see it on the fine cracks in the plaster of the walls, the threadbare carpet and the dirt shadow pictures that once hung in the half-dark corridors. '
*** THE ACCURACY Windows Speech Recognition ***
'I am a guest four, a journalist who reported on a sports event which so far hardly anyone has heard: the World Cup on Friday also. I am sitting at a small desk in a hotel room on the seafront overlooking the beach promenade in the Greek resort Kanama valley, the hotel is old, it looks a fine cracks in Potsdam Bender, the threadbare carpet and the dirt shadow images that once in The Hague dark corridors. '
Error:
four -> here
on Friday also -> Freediving
Resort Kanama Valley -> Destination Kalamata
Then -> you
a -> to
in Potsdam Bender -> in the plaster of the walls
Den Haag dark -> dim
Transitions - hung> gears
*** ACCURACY Dragon NaturallySpeaking ***
'I am a guest here, a journalist who reported on a sports event which so far hardly anyone has heard: the World Championship in freediving. I am sitting at a small desk in a hotel room on the seafront overlooking the beach promenade of the Greek Urlaubsortskalamata. The hotel is old, you can see it on the fine cracks in the plaster of the walls, the threadbare carpet and the dirt shadow images that were once hanging in the semi-dark corridors. '
Error:
Urlaubsortskalamata -> Destination Kalamata
dim -> dim
hanging -> hanging
As you can see, there was the Dragon NS actually fewer errors, and two of the three errors are only problems with word boundaries. I was also impressed that the program recognizes the name of a place (Kalamata), with the Windows speech recognition could not do anything. I was able to easily improve the accuracy of Windows Speech Recognition, after I read the section several times, but the results varied widely, and they never have the level of Dragon NS achieved. In contrast, consistently produces the same text when read again the paragraph with Dragon NS.
The Dragon NS software comes with a headset microphone, which is necessary in order to achieve a reliable speech recognition. This is not about the most comfortable headset I've ever worn, but it's not too bad. If you do not have one of these headsets: there is also a beautiful accessory for Skype calls and things like that.
When it comes to ease of use, I found the set up Windows Speech Recognition and easier to use than NS Dragon. I found Dragon NS every now and then something in error, especially when storing voice profiles. I also disliked that the program has forced me to register the product online in order to complete the installation. With regard to the performance, it is however a very impressive product.
Note that the home version of Dragon NS only recognize German. If you want to switch between German and English, you need the premium version.
