To Transcribe or Not To Transcribe

I have been testing automated transcription services for podcasts.  The results are interesting and inconsistent!  

At one point in the podcast, the speaker says "today the buyer often knows more about the particular model of car than a salesperson" 

This appears to have been a particularly difficult section for the transcription services to deal with.  

"Haha no more tissue. The cow says."

"Haha no more tissue. The cow says." is my favourite suggestion from the 4 services. 

Another two services felt that we were talking about tissues with suggestions of:

"tissue mortal car" and "tissue up on the call".

The fourth service took a different route altogether and suggests "'Tischler all car brands" - complete with capital T.

My point - if you are looking for a low cost fully automated transcription service, they have some way to go still.  To be fair, the file I used is my most podcast from last week (In Conversation at the Lead Gen Summit) and it is complicated. There are music introductions, multiple speakers and background noise.  However, this particular section is perfectly clear to the human ear. If you want to check for yourself, it is a 2:24 in the podcast (itunes.bridginggaps.uk).

One of my guests, Lindsay Milner (13:11 in the podcast) was apparently quite troublesome for one service in particular. Lindsay's voice is generally complimented, she is clear and easy to listen to.  When you listen to her audio, there are, admittedly, a few bits where her voice disappears into the noise. Not enough though, in my view, to add an entirely new person to our conversation (and renaming Lindsay)! The spelling of her name is forgivable as is transcribing 'Silvern Training' to 'Silver In Training'.

'Vincent Milner and Lindsey Hightower'

Interestingly, Lindsay also talks (according to the transcription) about batteries, a fishery and motorbikes!.

Should you try an automated transcription service?

It really depends on your objective and also what your source audio file contains.

What is great is that it gives you timestamps for a small outlay.  

It can give you a quick reminder of what was covered in the podcast rather than listening to the whole podcast again.

If you are going to consider it, I'd do a trial and see how well it works for you. If you are recording along in a quiet environment, it may well be very effective for you. Indeed when I have transcribed just myself - the result is much better.  However as soon as I added a second speaker in, this seems more challenging.

There are definitely some things that are easy to understand why the service gets it wrong - for example, Lindsay's company - Silvern Training, comes up as Silver In Training. That's fair! 

It can be amusing, for example, I'm still smiling at these:

'sharing the challenges and stories of feloniously Clow'

I am a little unclear on what "Geoff occultism" has to do with anything!

And I love:

Justice from the purple spider

I have to be fair on this one. Purple spider is correct. I was speaking to Jess Staniforth from Purple Spider Digital. However the idea of justice being handed out by a purple spider did make me giggle!