The nitty-gritty of how I record

I just responded in the LibriVox fora to a general question about “how does everyone record and edit?” — and thought that it might be good to put that detail here, as well as some info about my technical set-up. There’s a lot to say, it turns out! For reference, I only record my voice for audiobooks and podcasts. (Oh, and occasional short-form voice-over work … same settings / equipment throughout.)

I have a dedicated room, formally known as the hall cupboard (closet, for American readers) which is lined with noise-reducing foam tiles and material (mostly heavy velvet curtains.) It has a couple of shelves, one of which has an old-but-quiet laptop with its screen at a comfortable-to-read height. The lower one, at hand-height is for water, lipsalve and my keypad (more on that in a bit.)

My microphone is a Blue Yeti, it has its own floor stand which it is always attached to. It has sound card/encoding settings on board, so it just sends data to the laptop by USB and I don’t have to worry about pre-amps, computer sound cards and so on. Easy!

Before I record, I clean my teeth, get a glass of water and put on lipsalve (these all help reduce mouth noise for me.)

I record in Sony’s SoundForge to WAV. This takes about 75 mins for 60 mins of polished audio; however, I rarely record for more than half an hour at a time, followed by a 5-10 min break. SoundForge allows me to ‘punch in’ edits, that is, re-record immediately over the top of a mistake, and I have a gamer’s keypad set up with keyboard shortcuts to make this process easy. The keypad is old, cheap and effective! However, this way of recording extends the time spent in the booth significantly, so I don’t often work this way. I probably should … even though I’d still have to do a full listen/edit for overall smoothness, it’d still be quicker to process.

I then save a copy of the WAV file and noiseclean gently in Audacity. This step and all subsequent work is done on a file tagged -nc, (e.g. candy08-nc.wav.) That’s not what the final name of the file will be, just a shorthand that helps me know whereabouts in the process I am — very useful when I have several files at differing stages of the production process. My Audacity is version 1.3.12-beta (Unicode) and the Noise Removal settings I use are: 13 // 200 // 0.00. I always leave at least 10 secs quiet at the end of the recording, and use at least 2 secs as the sample of ‘room noise’ for the Noise Removal process. Happily, there’s very little background noise, just a gentle, constant hiss. Though, if the laptop’s fan has come on, these settings will also remove it without fuss as long as I include that steady fan-sound in the room noise sample.

Next I move to Adobe Soundbooth and compress gently. This is a relatively new step for me, but I was spending HOURS manually editing volume spikes and boosting quiet bits, so that’s much reduced by this step. Compression in very non-technical terms smooths out the volume, so the quieter sections are made louder and the loud spikes are quietened down. It took a LOT of faffing about (and reading lengthy explanations of exactly what compression IS, which I immediately forgot) to find good settings that work with my voice without changing the sound quality significantly. My Soundbooth is version 1.0 — a demo version that I’ve never upgraded. I would pay full price for the newest version if I could trust it to record, but it’s been unpredictable in that area in the past, and I don’t want to risk recording a lengthy passage and then have to rerecord because Soundbooth paused or glitched, both of which have happened in past trials. One day I’ll try out Audition … The (Advanced) compression settings that seem to work for me are: -3.0 dB, ratio 5.0, attack 1.0ms, release 100ms, output gain 1.0dB. Just to emphasise, I’m pretty good at listening to my own voice now, and these settings work well there, but aren’t at all guaranteed for anyone else! It’s definitely worth experimenting, if you have software with compression — just go GENTLY or you’ll squash the life out of your voice.

In non-LibriVox recordings, I leave the sound processing stage here; however, for LibriVox, I then reduce the whole file’s gain (volume) by 1 dB (on the Soundbooth scale, though I don’t know how that matches to Audacity etc.) Most LibriVox recordings are relatively quiet and especially if I’m working on a collaborative book, I don’t want to blast a listener’s eardrums too much compared to the previous reader! My settings are still on the louder end of many files I’ve seen, but not as far away from everyone else. The volume is increased during the compression stage, so really this is just putting it back to how it was before I did that smoothing out.

Here I’ll note that I do all my noise-cleaning before editing, so if it’s wrought havoc on the recording, I’ll pick that up as I listen through. Noise-cleaning at the end, just before MP3ing, caused me a few problems in the past which were only caught by PLs / end-listeners. However, there’s an equally good case for doing the processing after editing, in that if there IS a problem with the noise-cleaning, fixing it is just taking one step back in the process, not potentially having to do all the editing again too! My noise and noise removal settings are generally constant, so I seldom have problems with it and the first order works for me. Also, I have a decent pair of Sony MDR-XD200 headphones, which I do all my sound processing and editing with. It’s well worth listening through a few different pairs to find ones that let you hear as much as possible. There’s no way to know what end listeners will use, and it would be crazy to process the recording with a particular pair of headphones in mind … but you have to use SOMETHING so it might as well be something that sounds nice. It’s also very educational to try with a bad pair of earbuds, since many of your listeners will use those! Much of the detailed fussing I do over sound is made completely irrelevant there. 🙂 Helps me not to be obsessive about it! Also, the Sony ones are extremely comfortable — very important when I spend so long in them.

Once the file is processed and tidied up — which only takes about 5 mins, even for long files — I begin the editing. I use Soundbooth — it’s not the most obvious software, but I’m very used to it, and have lots of keyboard shortcuts set up to make it faster, e.g. C instead of Ctrl-C. I don’t make a noise/click when I make a mistake during recording (as some people do) because I’ll be listening all the way through anyway, editing as I go. I edit pauses (usually making them a bit longer), mistakes and annoying mouth-noises. This takes 4-6 hours for 1 hour of finished audio.

Finally I save a copy of the finished WAV file, then and then MP3 it. I use Soundbooth for this step too now: I used to use RazorLame, but then found I could HEAR a little sound degradation in the MP3 files produced that way. After checking every MP3 maker I could find, using exactly the same settings, I settled for using Soundbooth. Note: I couldn’t hear any difference IN Soundbooth (or Audacity) but in other players (RealPlayer & WinAmp) there was a definite difference. So that’s another way to check sound quality, using different software, or even a standalone MP3 player if you have one. Obviously I don’t do this for every recording, but it’s worth doing when I make a change in the process, just to be sure I’m still happy with the end result across a variety of output systems.

* * *

As I said in my forum post, the more one knows about recording, the lengthier the process seems to get. The more I learn, the more I want to apply to my recordings, and though I’m very happy with the output at the moment, there is a definite trade-off to be found between time and quality. This process has evolved significantly from the days when I recorded straight to MP3 using a headset mic, doing a quick edit before finishing, and I think my recordings over the past five years reflect that. Not all my improvements have been for the better (argh!!) and circumstances like house moves have changed the set-up too.