Which acoustic model do you use? Is it an acoustic model that you downloaded from VoxForge? What is your native language?

For example: let's say your native language is Dutch. The acoustic model for Dutch was made using less than seven hours of speech (right now we have collected seven hours, but the acoustic model was created some time ago).

This means that the acoustic model is not very good and your results will only be acceptable if your voice resembles the voice of one of the people who submitted speech (this is a bit simplified, but you get the gist). If you want to get better results, you should submit at least half an hour but perhaps even 2 or 3 hours of speech. Of course you should use the same microphone as you want to use later on (and the same soundcard). Then we can make a new acoustic model based on all the speech we have and in theory you will get better results.

This is one way how VoxForge can help you to train Sphinx. It is also the easiest way for you. However, it does require some work to make a new acoustic model, so you have to be a bit patient.

Of course, if Dutch is not your native tongue, you should not be submitting Dutch speech (unless your level approaches that of a native speaker).

Give us some more details regarding your setup and we can give you more advice.

--- (Edited on 8/23/2009 3:28 pm [GMT-0500] by Robin) ---