This is my longest post in ages. Created by voice recognition - probably got carried away - but hopefully, there are some useful tips in here.

------------------

Maximising your voice recognition accuracy

Background

Typing, even touch typing by an experienced typist rarely surpasses 80 words per minute for a maximum sustained speed. Most executives and senior management are hard-pressed to sustain a typing speed above 25 words per minute. The idea of voice recognition software has been around for a long time but, despite promised advances in both hardware and software capabilities, seems not to have penetrated the market to any great extent.

The situation has improved over recent years a significant number of users, after an initial dabble with the software, do not persist with the use of a voice recognition software package. Users have become sceptical of the accuracy claims of manufacturers. Even these claims that should be viewed with a sceptical eye. For example, 98% accuracy sounds very good until one realises that this means 20 errors per page. No human transcriptionist would get a job with that kind of error rate.

So how does the average user increase their success rate of these programs?

Equipment tips

Voice recognition software is still quite demanding on the hardware in your computer. A fast processor is helpful but generous amounts of RAM are more effective in producing accurate recognition. Because the computer has to work quite hard during this process, any internal distractions will affect accuracy. Therefore it is a good idea not to have too many open running applications when you are doing voice recognition. On slower computers, this also means shutting down some unneeded background tasks that may be running in your System Tray.

Choosing which voice recognition software application to use is now very easy these days. For the Windows platform, there are only two commonly available programs still on the market. Dragon NaturallySpeaking and IBM ViaVoice. These two programs are extremely similar in terms of features and performance. Both are easy to train and to use and are of similar price.

But there is one piece of hardware that is extremely critical to your success with accurate voice recognition. This sole factor is commonly overlooked and yet remains the biggest determinant in recognition accuracy. The microphone. When you think about it, this makes perfect sense as the computer will have much more success if it can clearly hear what you're saying, with a minimum of background noise, distortion, and a good range of frequency response. If you do a significant degree of voice recognition for your dictation, do yourself a favour and spend as much money as you can afford on a really high-quality microphone setup.

A recent innovation that has been very promising is the advent of microphones that feed their input directly into the computer via the USB port. Because the audio input circuitry on the average computer is extremely simple, even a high-end microphone fed into the simplistic hardware that comes with the average computer will not achieve optimum results. USB microphones that have a digital signal processor built into the dongle attached to the USB port have the benefit that the DSP can offload much of the signal processing from the computer's own brain. This job sharing results in increased accuracy.

Not all USB microphones perform equally well. Recent testing with a wide range of microphones showed that there was a clearly audible difference between one set up and the next. For those users who are planning to dictate a large amount of text input, it is well worth while researching this thoroughly. High-end microphones from companies such as Plantronics or Andrea are well worth the money.

Dictation tips

In days gone by, the software required that you pause briefly between each word. This is no longer the case, which is just as well because that style of dictation caused a serious amount of vocal fatigue. Indeed, current voice recognition software works best when you talk as normally as possible to the computer. Placing particular emphasis or stress on certain syllables or words often causes the computer to become more confused and less accurate in its recognition.

Speaking clearly has its advantages, but speaking naturally is much more important. Many users find that it works better not to constantly watch what the computer is writing and instead to only check the screen occasionally to see how the accuracy is going. Speaking in full phrases and with many words run together also helps with recognition accuracy because the software is highly dependent on the context of the surrounding words. It is for the same reason, that the software is much better at recognising fully expounded text such as in formal letters and reports, rather than in the abbreviated style used in short notes.

The software used to require considerable training time getting used to the nuances and vocal patterns of the particular user. Nowadays, this training time is very minimal, often requiring only five to 10 minutes of dictation of a short prepared text. However, training does not stop there. The software becomes much more accurate over time, if you tell it when it goes wrong. Using the built in voice recognition commands to correct the mistakes is a bit of a chore at first but pays off in the long run as the software becomes much more accurate at recognising what you're saying. It is well worthwhile putting in this time in the early days with the software.

Voice recognition software is still speaker dependent. This means that it is highly individualised to one particular user. This is not to say that the software cannot subsequently be used by another user, even on the same computer. It is simply a case of telling the software that the user has changed. However, it does rule out the use of voice recognition software for recording group discussions. There are some interesting research developments underway at present which look quite promising and being able to provide multispeaker input with reasonable degrees of accurate recognition. However, we are not there yet! In the meantime, try out this kind of software again. With a good microphone and a little bit of patience, you may soon find that you are able to quickly dictate documents with a minimum amount of correction and hassle.