|
Tips for Improving Speech Recognition Accuracy
Jon W. Wahrenberger
| The following tips, distilled from various sources, are intended only as a starting point. Additional guides on improving accuracy are listed at the end. |
1. Optimize Your Dictation Style: Despite our tendency to seek technical ways to improve recognition accuracy, far and away the most important key to optimizing accuracy with voice recognition software is the manner in which you dictate.
• Enunciate clearly
• Speak in a smooth manner, neither overly slowly or excessively rapidly. Speak in either sentences or phrases. Recognition is worst when speaking one word at a time.
• Maintain a relatively consistent pitch and tone when dictating
• To hear some examples of good dictation and bad dictation, look at the examples from emicrophones.com ( http://www.emicrophones.com/articles/index.asp )
2. Optimize Your System
• Don't run unnecessary programs (will compete with available RAM)
• Run disk defragmenter periodically (programs > accessories > system tools > disk defragmenter). This can be somewhat time-consuming, so don't start it in the middle of the day!
• Be sure your computer has adequate RAM; although 512 MB is considered the minimum, a full 1 Gb or more may be optimal. If you are using Windows Vista, double these figures!
• If you notice recognition accuracy deteriorate during the day, consider restarting your computer to clear your RAM
• Use the best microphone you can afford
• If you are using a laptop computer and are not getting great accuracy, consider bypassing your integrated "sound card" by using a "USB sound adapter". This is a device into which you plug your microphone and which in turn is plugged into a USB port on your computer. The USB pod digitalizes the analog signal from your microphone and bypasses the internal sound card on your computer. Two commonly sold USB pods are made by Andrea and VXI. These can both be purchased here (view accessories) and other microphone dealers.
3. Optimize Your Software
• Run the "audio set-up wizard" (tools > accuracy center > check your audio settings) when making a major change in dictating environment or when you notice a deterioration in recognition accuracy Train Software. (The audio set up wizard can also be opened simply by saying "Check Audio".)
• Let Dragon learn from its mistakes (read the appropriate chapter in the User's Manual that came with your version of dragon) - but do not utilize "training" methods if the mistake occurred because you enunciated poorly.
• Before training an apparent "error" you should listen to the dictation either by setting dragon to automatically "automatic playback on correction" (under the tools > options > correction menu) or by choosing "Play that Back" from the correction menu.
• If your enunciation was clear, you can reasonably correct the dictation.
• If your enunciation was not clear, you should simply re-dictate the wrong word or phrase, otherwise the poor enunciation will be stored with your correction and acoustic model degraded.
- After establishing a good working user profile (typically after a week or so or regular dictation and corrections), stop saving your user files regularly. Rather, save user file only after making major corrections or additions to your vocabulary.
4. Other on-line guides to improving accuracy:
Key Steps to High Speech Recognition Accuracy - Douglas Durham, with help from Martin Markoe, Susan Fulton and Judy Evans
Tips for accuracy with Dragon NaturallySpeaking - by Peter Maddern at Speech Empowered Computing
|
