16 bit will provide enough dynamic range for just about any audio recording (if you have your amp turned up loud enough to hear the the bottom few bits, then the top few bits will rattle your windows and set off car alarms in the street), but that is assuming that everything else is perfect. When recording, nothing is ever perfect (find me a singer that can consistently hold a peak level +/- 6dB).
Ampex was far more forgiving of peak overload - tape saturation, so long as it is not too extreme, will give quite a musical compression - but had a pretty high inherent noise floor without fancy compression/decompression/pre-emphasis/de-emphasis.
For digital recording, 24/32 bit provides a ridiculously high dynamic range (in most cases far greater than the rest of the recording equipment), but means that you can leave a lot of headroom with no fear of running out of bits at the bottom end. For high quality live recording, I prefer to use as little compression (or other processing) as possible during the recording - usually just a hardware peak limiter to prevent an otherwise good take from being wrecked by a spurious peak.
Although a digital format of 24 bit or higher has enormous dynamic range, the dynamic range of the rest of the system still needs to be considered. Different microphones have different sensitivities and should ideally be kept well within their operating range. Too high a signal and they will compress, then distort, then start doing very nasty things (the sound of condenser plates arcing is emotionally disturbing). Too low a signal and you’re running into microphone self-noise. Ideally you would choose your microphone according to what you are recording. In practice and with a limited budget, good compromises can usually be found by careful microphone placement, but there are limits (you can’t get a “close mic’d” drum sound with a shotgun microphone, and you can’t get a clean recording of speech from 50 feet with a dynamic vocal microphone).
Back to the Sony pcm D50. I’ve not used one, but I do use a much cheaper recorder of a similar type (Zoom H2) and have had excellent results from it. I would expect the Sony (at 3 x the price) to be better.
The Zoom H2 has 2 recording level controls - there is a simple H/M/L sensitivity switch, and a 0-127 level adjustment. The sensitivity switch is the important one. The switch changes the loading on the microphones and adjusts their sensitivity, whereas the the level adjustment scales the A/D conversion much in the same way as amplifying in Audacity. An analogy from the world of digital cameras is that the switch is like an optical zoom, whereas the level adjustment is like a digital zoom. While the digital zoom on a camera can make the picture bigger, it does so by looking at less pixels (just the ones from the middle of the picture) and spreading them out, so the picture is bigger, but the image quality goes down - just the same as cropping the picture in PhotoShop.
Using the Zoom H2 fairly close will usually produce much cleaner recordings than recording at distance, as the amount of extraneous noise pollution and room reverb is generally much greater than it sounds to your ears. Human hearing has a remarkable way of “focussing in” on what you are listening to and filtering out unwanted noise - microphones do not do this. I usually have the recording level set to 100 and leave it at that, and use the level switch to get the recording level in a reasonable (not clipping) range for close up recording. If the recording level goes too high even with the microphone set to the least sensitive setting, then I move the microphone back a bit. The Zoom does not have an internal limiter, just an AGC which is thankfully off by default.
For recording speech or vocals I use a fine mesh pop filter (not the foam wind-shield provided) and (ideally) place the microphone about 30cm from the singers mouth. The Zoom H2 also has a configurable pickup pattern (90 degrees front, 120 degrees back, omni) to choose from. When recording in an acoustically nice space, the 120 degree or omni-directional are nice to use. The 90 degree (cardioid) is better for isolating the recording source.
The file format settings go all the way from 48kbps MP3 (which will give about 4 days of low quality recording on a 2GB flash card) up to 24/96 WAV which will give about an hour at very high quality (though this requires a good, fast flash card to handle the data rate).
If I am going to be putting the recording onto CD I will record at 24bit 44.1kHz (CDs use 44.1kHz), then transfer the file over to Audacity via USB (the zoom can be connected to a computer so as to appear as an external disk). As the file is 24 bit, I can adjust the volume virtually losslessly, but Audacity will apply dither when it processes the sound (processing is done at 32 bit) so I will convert to 32bit (this conversion is lossless). If there will only be editing and no processing, the format may be left at 24bit.
If there are any significant high peaks, I will probably apply a bit of compression using the SC4 compressor plug-in (rather than Chris’s Dynamic compressor plug-in).
While Chris’s Dynamic Compressor is excellent, it affects virtually the entire dynamic range, and this is usually not what I want. Chris’s Dynamic Compressor is brilliant for levelling out (reducing) dynamics, and so is ideal for making CDs that you want to listen to in noisy environments (such as in a car), but for listening to music I would usually like to be able to hear the dynamics as they are an essential expressive part of the music. All I want to do is to drop the peaks a little so that I can get a good (not too quiet) level on the CD without clipping, and for this the SC4 compressor is better.
After editing, I will Amplify to around -0.3dB and Export as 16bit WAV for burning to CD (allowing Audacity to apply dither). There are a choice of dither settings that can be argued about, but the default setting is good. If you want to be totally obsessive about sound quality, Audacity can be built from the source-code to use “libsamplerate” instead of the default “libresample”. This option should not be used in conjunction with VST plug-in support.