Now that I've got a means of recording that does not constantly "glitch," I can turn my attention a little more to processing the raw voice. It would be nice to think that with the right input I could just perhaps normalize it a little bit and not worry about compression, but it appears that is just a fantasy. Both the Snowball and the Logitech headset seem to give me a dry RMS level of about -35 dB, which is quite low. Just normalizing that introduces a lot of noise, and even though the starting level is quite low, without compressing the peaks at least a bit, you can't get the RMS level under 20 dB. The Snowball is off-line for the moment, so I'm talking about the Logitech USB headset.
I spent several hours last night working on an effects chain that would produce a result that I was happy with. Here is the chain that I came up with, in order:
I have a moderate setting on the compressor, with downward expansion engaged, again to a modest degree; I'm not quite treating it as a gate. I previously spent quite a bit of time messing around with the Floorfish to try to remove background noise, but it wasn't coming together for me. In particular, the Floorfish seemed to have trouble with threshold around the low incoming level of the dry signal. With no numeric display on the threshold control, I was operating entirely by ear, and I didn't like what I heard. It seemed like the processor just needed a lot more signal to work with, but I wasn't sure how to get a reasonable pre-gain.
There is probably a way, but I found that the Apple dynamics processor has a combined compressor and downward expander which acts as a decent noise remover, and I was able to get a better-sounding signal right away, and then spend some time tweaking it.
I still have a few issues with it, though: the meter only goes one way, showing the downward expansion but not what it is doing when the compression kicks in. Also, it does not have a "ratio" control, but instead a "head room" control, measured in dB. I feel that I understand "ratio" and "threshold" in compression, but I can't quite figure out or hear what the different settings of the headroom control really did. Is this the additional dB of dynamic range your signal is allowed to fill up once passing the threshold? Or the number of dB you want to remain unused after your signal has reached its maximum volume? I'm not really sure.
It also has only a single master gain control, rather than an input gain and make-up gain. This resulted in a lot of time spent tweaking, because when you're working with compressors, the results can be counter-intuitive (for example, increasing the gain can result in the compressor getting hit harder, which will actually give you a lower RMS level, as more of your signal passes the compression threshold).
I'm using a slight variation on the "male voice" preset, which takes some of the hissiness out of the Logitech mic. This plugin is fairly subtle, and if you set it to take too much out, my voice sounds like I have a lisp. At a relatively low setting it definitely helps. I find it difficult to figure out the most effective frequency (since, again, the "fish fillets" don't display their settings numerically). And, I still get a few pops; I can't really put a pop filter on the headset mic, and I don't want to change the boom position because it seems to be in the "sweet spot" for tone, so I'll live with it for now. (It seems to be difficult, if not impossible, to take out plosives using EQ -- they cover a very broad band.)
3. Graphic EQ
Apple's EQ set to 31-bands. I have rolled off the very high and very low ends, and set a slight boost in some of the more flattering bass frequencies. In addition, I've taken out a little notch around the more "nasal" sounds my voice produces, and another notch to tame the sibilance frequencies just a little bit more (some of this is, I think, resonance in the headset mic's little plastic boom). The rule here is generally to prefer cutting to boosting, which I've done, and to be subtle. I did a lot of A/B comparison until I was satisified, but I reserve the right to tweak it more later!
4. Peak Limiter
I'm using Apple's plug-in. I can't really figure out what this thing is doing. It is very hard to read the controls and the "limiting amount" slider doesn't seem to change anything -- I think that is a bug. But if it is in the chain, it does indeed seem to clean up the very highest peaks and reduce the number of clips.
I'm enclosing a screen shot of the plugin configuration:
And here are some large (about 45 MiB) dry and wet AIFF files (note: these are un-edited, raw readings!)
I'm pretty happy with the result -- I think this cleans up the output of the inexpensive Logitech USB headset just about as much as it can be cleaned up -- but I would welcome any suggestions -- particularly in how to deal with the AUDynamicsProcessor "head room" setting and the peak limiter.