A lot of the investigation that has been done so far in my institution on subtitling has focussed on the technical side – which transcription services can give the best automated results etc. Whilst this is useful insight, I don’t think it gets to the heart of the biggest issue in subtitling at scale. Whether an automated solution is 70% accurate or 80% is to some extent splitting hairs, because what is certain in both cases is that human mediation is required to get an accurate transcript.
Decisions on preferred automation technology are pretty much straight down the line decisions about cost versus quality and that’s the sort of thing that’s pretty easy to do. Making decisions about human labour is much trickier though. If we are going to do subtitling at any kind of scale, we are going to spend significant amounts of public money on human mediated transcription because that’s the most expensive bit.
Many of the large subtitling companies obscure human labour behind the veneer of their platform; you buy human labour as a service via the platform, complete with an advertised price per minute and an SLA (100% accurate transcription with 48 hour turnaround etc). People can’t be optimised to run faster in the way that machines can though. There are finite limits here. So heavy discounts or cheap prices should translate in our minds immediately to questions about where this work is being done, and what people are being paid to do it.
Looking at various sources, the suggestion is that the industry standard is around an hour to transcribe 10-15 minutes of audio. If we assume a worst case of an hour per 10 minutes of audio, then an hour of spoken word will take 6 hours to transcribe.
The Real Living Wage in the UK is £9 per hour. On that basis an hour of audio should cost £36 – £54 to transcribe, which breaks down to about £0.60 – £0.90 per minute.
The legal minimum wage in the UK is £7.83. On the best case basis that it takes 4 hours to transcribe an hour of audio (an hour per 15 minutes) that works out at £0.52 per minute.
Anything below that we should be asking very serious questions about where the work gets done, who does it, and what kinds of conditions they work in.
One of the ideas we have in our subtitling project is to pay our own students to do the work to complete transcriptions. We were recently advised that subtitling and transcription work is often the kind of work that is particularly accessible to disabled students, who otherwise might struggle to find suitable work. It is flexible work and can be done from home. That kind of work could also very easily support students who are parents, carers, or from less affluent backgrounds where money for travel expenses is often prohibitive both for study and work.
Then I spotted this Tweet from @Typeology
We are a workers' co-op founded by disabled people who make audio transcripts. If you have a podcast and funding for transcription (we know not everyone does), contact us! https://t.co/laHoN6V6Ue
— TypeOlogy Co-operative (@TypeOlogyCoop) November 6, 2018