Machine transcription isn’t ready for prime time
As you may know, I host the A11y Rules Podcast and its offshoot, the A11y Rule Soundbites. And I provide transcripts with every show, at the time of publication. Transcripts are expensive to generate if you use human transcriptionists (note, this is not a comment on the value of their work!). There is a lot of talk about using automated transcription services. Recently, the folks from Otter.ai have been pushing for me to give their service another go. And so I did.
The results are underwhelming. The time it took to rectify nullifies the time saved by using automated transcription services. There were a lot of small mistakes, like “explained” instead of “explain”. But there were also some more significant problems.
One of the things it was not able to reliably determine is who was speaking. This lead to sentences assigned to one person instead of another. My guest for this show is from the United Kingdom and has a British accent, although not a thick one. His accent and his voice are quite different from mine.
There were also some more significant mistakes that changed the meaning of sentences. Or that made sentences incomprehensible.
But check it out for yourself. Here are two versions of the first paragraph. The rest of the transcript continues along the same lines. The first version was transcribed by a human. The second was transcribed by otter.ai. I highlighted the errors in the second paragraph by using emphasis and red.
Human transcription
Hi, I’m Nic Steenhout. And you’re listening to the Accessibility Rules Soundbite, a series of short podcasts where people with disabilities explain their impairments, and what barriers they encounter on the web. Thanks to Tenon for sponsoring the transcript for this episode. Tenon provides accessibility as a service. They offer testing, training, and tooling to help fix accessibility fast. This week I’m speaking to Jean Luc Sorak. JL is a software engineer at Holiday Extras in Kent, United Kingdom. Thanks for talking to me. JL. How are you?
Machine transcription
Hi, I’m Nic Steenhout. And you’re listening to the accessibility rules Sunday, a series of short podcasts where people with disabilities explained their impairments, and what barriers they encounter on the web. Thanks to tenant for sponsoring the transcript for this episode 10 and provides accessibility as a service. They offer testing, training, and tooling to help fix accessibility fast. This week I’m speaking to john Rick sarek. Jail is a software engineer at Holiday Extras in Kent, United Kingdom. Thanks for talking to me. Jail. How are you? Very good. Thanks. Yeah. Thank you for having me on. My pleasure. So let me ask you what is your disability or your impairment? So I have dyslexia and dyspraxia they often come as a pair.