To record sound samples, hook up the CIMEL microphone to a computer with Cool Edit. (We always use the same microphone to get a consistent sound.) Place the mic on a stand so that it is positioned in a way which will allow the speaker to talk "over" the mic, rather than directly into it.
All wav files should be recorded at 22KHz, Mono, 16 Bits, as per the standards written up on the Cimel website in the Guidelines document.
Each file should be normalized to about 150% to get a good signal-to-noise ratio. You can vary the exact percentage a little bit (typically less than +/- 20% variation) if you need to adjust the loudness of the sample to make them all consistent. A good rule-of-thumb to figure this out is to let a few of the highest peaks clip slightly over the horizontal lines displayed at the top and bottom of the Cool Edit edit window. Don't stray too far from 150% normalization, otherwise the levels of background hiss will vary a bit too dramatically from sample to sample. If you find that a sample simply isn't loud enough, or is too loud, adjust the distance of the microphone to the speaker and re-record.
Don't leave too much empty space at either the beginning or ending of the sample. Trim out everything but the necessary portions.
As far as mp3 compression, any compressor that uses the FhG codec can be acceptable. All files should be compressed with a bitrate of 56000, and make sure that the resulting mp3 stays in mono. Batch encoders that can be used include: mp3enc, using l3enc, and l3encwin, using l3enc. l3encwin is available at download.com. If you use l3encwin, a good thing to do is remember to uncheck the "Auto-delete wav files" option. Cool Edit 2000 (not Pro) can also save files directly into an mp3 format, using FhG, which is another option if you compress each file as you go.
It's also a good idea to cue up a whole bunch of files in screen order on Winamp or any other wav/mp3 player and listen to make sure that the sound quality is consistent through out them all. Occassionally, you'll find that you might need to re-record or re-process the occassional file, because the sound levels might be off.
And that's it!