Recording an audiobook
Perhaps I've been listening to too many Doctor Who audios, the CDs given away on the front of the Radio Times a couple of months ago, David Tennant himself reading The Feast of the Drowned and the Stone Rose, and of course the Big Finish radio series on BBC7, but I suddenly had a yearning to produce my fan novel, Michela's Door as an audiobook.
You might have thought that actually finishing it (I've completed 12 chapters, which is the first of three parts) might have been a higher priority, but apparently not. I installed Audacity, a free audio editor which has a reputation for being relatively easy to use, yet powerful enough to actually get stuff done - this is ideal, because I don't have any plans to become a sound engineer, I just want to be able to get things done without too much faffing.
After a bit of tweaking, I was able to get something recorded:
- the microphone level was set to 0 by default
- audio recording was set to mono, which produced a terrible noise
- the bit rate was set too low, I reset it to 96000
I did the first demo, mixing in the sound of a mix from Whomix, the Jah Humphreys Dub Explosion, which I thought would go really well with an audio book. I asked Tardy, a big Who audio fan for comments, and he agreed that, er, yes, I speak far too fast (this isn't a surprise. I do. Always have done).
I followed another suggestion of his, to split the recording into chunks. This has several advantages: if you need to rerecord one bit, it's far easier, and because you are stopping and starting, it's easier to slow yourself down (I tend to find that the longer I read, the more I forget to slow down.) I managed to get down from >200 wpm to <190 wpm, still rather fast, but a big improvement.
I then found that my poor old laptop didn't like dealing with all these tracks, if I tried to play the mix the hard disk would start thrashing. I guessed that the file size (about 780Mb in total) had to be stored in memory all at the same time and it was unhappy.
I started to record again at a lower bitrate: 44100 was too low (it produced distorted recordings), but a rate of 48000 seems to be fine.
Doing little tracks is great, you can work on one at a time: if you substitute, then you have to faff a little getting all the subsequent tracks aligned properly (there is a shortcut in the Project menu to make this a bit easier, though I'm still finding it a bit clumsy). Also, it's easier to tell how long they are: I have a spreadsheet open where I calculate my words-per-minute rate. I'm aiming for 180, but I'll re-record if I get over 185 for any one chunk.
Actually, most chunks have been rerecorded several times. There might be a click I couldn't get rid of, I might mumble a word, or my mouth might be too dry, or too full of saliva.
I've also found that by reading out loud, I've found lots of tweaks to make to the story text itself: places where I've repeated the same word in quick succession, or where it's impossible to keep the flow with the words I used, or where I realise that a sentence doesn't really make sense the way it's written. I'd recommend reading your work out loud (ideally recording it so you can hear it back again) even if you're not planning to create an audiobook, just for the additional insight you get into how to improve it.
I've discovered that if you have *too many* audio tracks, the program will thrash to a halt when you try to playback. I think it's because it has to mix them all together before it can play: given that they are mainly offset in sequence, there are usually no more than 2 tracks that could be playing at any one time, so I'd have hoped that it would optimize this a bit. But apparently not. For my reading of chapter 6, which I'd split into about 20 chunks, I went through again, running "Quick Mix" on 2-3 chunks to join them together again.
Another disadvantage of chunks: when you play them together, you realise that there are differences in the sound quality between them. Given that I've not been playing with the recording settings in the mean time, this is probably because of
- dryness of mouth, how much I've been talking by that point in the day
- how close the microphone (on a flexible arm) ended up to my mouth
I think that there's not much I can do about this now: I don't think that rerecording everything makes sense, after all, if I'm doing it in chunks, how would I know that it wouldn't happen again? Maybe I'd need more than one version of each chunk and then choose the ones that work best.
I imagine that a trained audio engineer would be able to filter these chunks in some way to minimize the difference in sound quality (or handle the recording so as to minimize these differences).
Audacity by default doesn't know how to export to MP3. I exported to OGG and then used ffmpeg to convert to MP3, but that was producing quite a low quality file. I tried exporting to a WAV file (which is lossless) and converting that, but this time the quality was worse, with a strange distorted echo.
In the end, I pointed Audacity at the right libmame libraries, and the export to MP3 is fine, though, annoyingly, it asks for the ID3 tags every time you run the export, rather than remembering them.
Recording an audiobook is not hard, but it takes a little bit of faffing and quite a lot of time. I think that to produce just under 11 minutes of finished audio it has taken me about 7 hours of recording, editing, rewriting, and assorted faffing. Obviously, in that time I've learnt a lot about the process, and about how to use Audacity. From reading some information about professional audio book recording, I believe that a team made of "Voice talent", producer, and sound engineer would take about that many man-hours to do about an hour of recording. Hopefully I will get the next 10 minutes done slightly quicker...
Current Mood: accomplished 
tahwail