Speech Recognition That You Can Actually Work With


I happen to have covered the speech recognition arena for many years, and there are few technology product categories that have remained in “almost there” mode for as many years as this one. For the longest time, speech recognition software was capable of about 80 percent accuracy for intensive tasks such as taking dictation, but that just isn’t quite good enough. Now, however, I have found one speech recognition tool that helps me on an almost daily basis, and can probably help most web workers: the Sony ICDMX20DR9 Digital Voice Recorder with Dragon Naturally Speaking Software.

With this recorder and the accompanying software, I still can’t get perfect transcriptions of phone interviews I do (I do a lot of these), but the accuracy is remarkably good.

For most speakers on the other end of the line, the Sony recorder and Dragon combine for about 90 percent accuracy. If the person I’m speaking to has unusual inflections or an accent, the accuracy rate is lower and I tend not to use it, but I’ve gotten to the point where I can tell in advance about how accurate the recognition will be.

Dragon’s software is recognized by many people as the best commercially available speech recognition software, and I’m glad to see that ScanSoft is continuing to develop it, after it languished for a few years. Note that whenever you use speech recognition software, you want to train the software in advance.

If you do a lot of transcriptions of interviews and the like, at about $270 online, this combination of a digital recorder and software is a real time-saver. It won’t work all the time for every person who is speaking, but it works well enough to be very useful.

Do you use speech recognition software?



I would like to hear more. I am dictating this using NaturallySpeaking 9, but I had assumed that my high accuracy was due to the fact that I had trained the software myself. Anecdotally, I have noticed that when I demonstrate the software to other people, they often achieve a fairly impressive level of accuracy with out having trained the machine at all.

How do you identify the speaker to Dragon Naturally Speaking when you do one of these interviews? Do you simply let it and assume that it is listening to you?


In my experience the results/accuracy you achieve with Dragon NaturallySpeaking when using sound from a digital recorder depend heavy on the quality of the mobile device that you use. I tend to recommend the Olympus Pro range of recorders such as the DS-4000 and DS-5000. Typically most users get around 90%-95% accuracy with these devices as long as they are good speakers.

Comments are closed.