Monday, 11 May 2020

The transcripts time factor

When I did my Master's degree a number of years ago, I went to what I called "Transcription Hell" (I have written other posts which you can read here). You see, when I applied for my research ethics, I thought idly about paying to get someone to transcribe my recorded data, but with the opportunity cost of the Master's, and the naivete of "how hard can it be?", I decided that I should be able to push on and do it myself.

Ha, ha: what a flaming nightmare THAT was. To begin with, I had to keep going back and back and back over what I had recorded to actually hear what was said, Play, pause. Rewind. Play, pause. Rewind. Play, pause.

I couldn't type fast enough to get the words as they were said. After a few days of being demoralised, I hit on the brainwave of getting some transcription software, and cheerfully bought Dragon NaturallySpeaking, fondly believing that I could just play the recordings to Dragon and it would hear it and transcribe it for me. Would it what!

Instead I still had to listen to the data, and just speak it aloud for Dragon to hear my voice. It also took me probably 10 hours of work to train Dragon to hear a Kiwi accent, and it still could not hear "umm" by the end of my hundreds of hours of transcription time.

"Eeek: 'Hundreds of hours'?", I hear you shriek? Yes. I worked out that there was a factor of 60 in transcribing data. I found that it took me a minute to process for every second of recording. Thus each hour of data took roughly sixty hours. I had 20 hours of recordings, so that added up to about 1200 hours. It took me five months to get it done, in and amongst other things that I had on my plate. 

Yes, transcription hell.  

So when I read Getchen McCulloch's book, "Because Internet", recently, I was most gratified to read that "[i]t takes about an hour of skilled human work per minute of audio recording to get speech into a transcript usable for linguistic analysis: to transcribe the overall gist, to go back and add detailed phonetic information, to extract parts and analyze their acoustic frequencies or sentence structure." (2019)*.

Ah, yes! It appears that the ratio I found for transcription was pretty bang on. 

In tackling a PhD, one of the first things I did was to tee up a transcription service. There is no way I am going to transcription hell again.


No comments :

Post a Comment

Thanks for your feedback. The elves will post it shortly.