This is quite a challenge for
any audio program - the signal level is very low and the noise very high - it's not going to be possible to make it sound perfect.
The first thing I did, since it is a mono recording, was to convert it to mono(split track > delete one track > Set the remaining track to "mono") and normalise to 0dB ("Edit->Amplify").
So here is the mono track that I worked on:
http://easyspacepro.com/audacity/testsample.wav (1.6 MB)
The settings that I quote below are not optimal settings, they are my "first attempt" settings.
Noise reduction in Audacity 1.3.4 is quite different from v.1.2.6, so I tried both. Overall I think 1.3.4 performed slightly better, but it's always a trade off between quality, noise and artifacts.
The plug-ins that I used are available in the plug-in packs linked from the main Audacity website.
Audacity 1.2.6
The "noise" element is broadband (high and low frequencies), but the audio (speech) is mostly mid range, so the first thing was to eliminate much of the noise by simple filtering:
1) Low pass filter at 2.5 kHz 24 dB per octave
2) High pass filter 400 Hz - 24 dB per octave
3) DC offset removal (this will generally improve the performance of noise removal processes, but in this case was probably not necessary)
Listening to the noise, you can hear a low pitched hum and a crackly hiss. Noise removal will be more effective if we can deal with these separately.
4) Duplicate track
5) Low pass filter track 1 at 1.2 kHz - 24 dB per octave
6) High pass filter track 1 at 1.2 kHz - 24 dB per octave
7) Select a section of noise from track 1, open the "Noise Removal" effect and click on "Get Noise Profile".
8) Select all of track 1, open the "Noise Removal" effect and set the "step 2" amount to about 15% (near to the last "s" of "Less")
9) Click on "Remove Noise".
10) Repeat steps 7-9 for track 2
Listening to each of these tracks individually (select "solo") I could hear audio artifacts (bubbly metallic type sound) on both tracks - the artifacts were mostly mid frequencies, hence step 11)
11) By running steps 5) and 6) again, the artifacts could be reduced .
12) Run "Project -> Quick Mix" to combine the tracks back together
13) Amplify to 0dB
14) Using a "Gate" effect (LADSPA plug-in) it was possible to reduce the noise between the talking.
The settings I used on the "Gate" were:
LF = min
HF = max
Threshold = -20dB
Attack = 50 ms
Hold = 90 ms
Decay = 50 ms
Range = -90
Output = 0
The final result is here:
http://easyspacepro.com/audacity/clean%201.2.6.wav (1.6 MB)
Audacity 1.3.4
Basically the same technique, just a few small changes.
I set the project to 48000 Hz and worked in 32 bit float, converting back to 44.1/16 at the end.
In the initial "Low pass filter" (step 2) I ran it at 3 kHz for a bit more
clarity to the voice.
The Noise reduction settings in steps 7) to 10) were approximately as follows:
Noise Reduction = 40 dB
Frequency Smoothing = 300 Hz
Attack/Decay time = 0.2
I didn't bother with the second filtering stage (step 11) as there were far less "artifacts"
The result before gating is here:
http://easyspacepro.com/audacity/clean%201.3.4.wav (1.6 MB)
and after gating:
http://easyspacepro.com/audacity/clean% ... 0gated.wav (1.6 MB)
As I said, the exact settings above are not "optimal" but they should give you an idea. After a bit of tweaking, I got this using Audacity 1.3.4
http://easyspacepro.com/audacity/clean% ... 0gated.wav (1.6 MB)
Just to give an idea of how difficult it is for a program to deal with audio this bad, this is the result using professional quality (expensive) software - you will notice that it is still far from perfect:
http://easyspacepro.com/audacity/clean%20pro.wav (1.6 MB)
)