Cali 2 NL

It may be too late for some of my fellow CRG-program buddies (some of which are much further along with their theses than I), but I figured that my process for transcribing interviews could be useful for someone else out there.

High-Quality Digital Recorder
I had no audio-recorder of any kind before I began this process, so I started doing a bit of research on the Internet and by talking with colleagues. Everyone told me not to skimp on the quality of the recorder, since you don’t know what kind of control you’ll have over your interview environment and some recorders are better at dealing with ambient noises than others.

I went ahead an purchased a Sony ICDUX70 Digital Voice Recorder for $60.00 online, and was very happy with it. It records directly to mp3, holds 12+ hours on the highest quality setting, and has a USB port for downloading with no special software required (works great on my Mac).

I’ve had two interviews in noisy areas, one was at a bar with not-so-quiet music and not-so-quiet guests, and was impressed by how audible the interviews came out.

Take Lots of Notes
Hey, you’re recording this with your fancy new digital recorder, so no use in taking copious notes, right? Wrong. Whenever your interviewee references a new name (organization or individual), acronym, or jargon, write it down. At the end of their response (or the end of the interview if you don’t want to break the flow), go over those names/terms and make sure you’ve got them right. Humans are agreeable creatures, and sometimes I find myself thinking that I’ve understood a term or reference, only to find out that I’m not 100% sure during the transcription process. This step saves a lot of time and prevents you from having to write a potentially embarrassing follow up email.

Voice Recognition Software
Here’s where I got a bit creative. Speech-recognition software has come a long way since my last encounter with it back in 2000, but it still can’t accurately understand a voice that it hasn’t been trained to recognize. So, if you’re hoping that you can just run your recorded interview through a computer and receive a relatively accurate transcript, you may have to wait another decade.

That doesn’t mean that it can’t be pretty useful, though. I already had a copy of MacSpeech Dictate (equivilent of Dragon NaturallySpeaking for Windows), and I invested about 30 minutes of training-time before the software could do a pretty good job at understanding what I was saying. This kind of software isn’t cheap ($99-199), and it works better with some voices than others (for example, accented speakers). I really would recommend playing around with a demo or an internet copy to see if this will work for you in the long run, before investing in something which may not be that useful for your specific situation.

Here’s my process:

• Spend some time training your voice software (and yourself) until you get to a point where it can understand a good deal of a practice text.
• Copy mp3 interviews onto your computer.
• Fire up an mp3 player capable of adjusting the play rate without changing the pitch (QuickTime will do), and set it to half speed playback.
• Plug in your headphones and make sure you can clearly hear your recording.
• Start your voice software.
• Playback your interview and speak out-loud what you hear through your headphones. This takes a bit of practice but at half-speed playback it’s not too big of a challenge. Don’t worry too much about errors, your job is to get as many of the words as possible into a text-editor.
• Close your voice software, and go back and format the messy document you’ve made as best you can.
• Now play back your interview again (possibly at normal speed), and edit your transcript as you normally would, pausing after each sentence and rewinding during difficult moments.

This process is far from perfect, but I’ve found that listen-and-repeat method can usually get a good chunk of the data into text format which I can then edit. Editing an incomplete transcript, of course, is always easier than doing the whole thing from scratch. I’ve been able to use this method to get transcribing down to a 4:1 or sometimes 3:1 ratio of transcription-time to interview time, which isn’t too bad.