Audacity has a number of editing features to taylor your final wav to tastes.
Normalize or Amplify
Hardlimit (any enthusiastic drummer / door slam / other thing that was loud and would limit the average loudness of everything else can be squashed down / tamed.)
Multiband EQ (sometimes there wasn't enough bass, or some annoying sound in a particular range that you can adjust to more tollerable levels.)
Select the portion of audio you want to edit in Audacity then:
Effects -> Amplify
Effects -> Normalize
Effects -> 151 to 165 -> Hard Limit
Effects -> 181 to 195 -> Multiband EQ
i.e. everything doesn't have to be perfect at the point of capture to get a good result. There's also ways to do the same things with sox and other software. Unfortunately a lot of them are not included by default and might have weird licensing or regional legalities that prevent them from ever being included in a lot of distros.
One should also note that what you see as 0.5 in the audacity waveform is really -5dB. If you were to hardlimit to -5dB, it would fall about where that point is. Or amplify to -5dB (which might actually do the opposite).
-----
The quirky issue with ffmpeg and realtime capture from a script like this:
Code:
#!/bin/sh
ffmpeg -f oss -ac 2 -ar 44100 -i /dev/dsp -t 00:00:30.000 \
-acodec pcm_s16le -ac 2 -ar 44100 -y ttt_ffmpeg.wav &
sleep 1
for PAUSE in 0 1 2 3 4 5; do sleep $PAUSE; echo "test test test" |
festival --tts; done
sleep 20
# ----------
arecord -f cd -d 30 ttt_arecord.wav &
sleep 1
for PAUSE in 0 1 2 3 4 5; do sleep $PAUSE; echo "test test test" |
festival --tts; done
sleep 1
As seen in audacity:
$ audacity ttt_arecord.wav
File->Import->Audio [ttt_ffmpeg.wav]
http://home.earthlink.net/~shadow_7/ttt_result.gif
Technically the only difference between the two results should be some latency of when the applications loads and launches. Which should be in the neighborhood of less than 1 second. But as you can see in that screen capture, the difference is a lot more extreme. In ffmpegs defense I guess, ffmpeg does peek my system resources on my 32 bit 2GHz laptop circa 2006. In combination with the script and festival and other stuff since I ran the test in X. arecord has a bit more elbow room. I should note that speaker output was consistent, despite the resulting outputs.
In short ffmpeg doesn't work for me at the point of capture. Not to imply that it doesn't work at all. Or doesn't have a fix in the most recent cvs/svn/git. But my version (early 2010 / svn-r21496) on my gear (2006 laptop), not quite a result that I'm happy with. You're better off with arecord or audacity or ardour IMO.