Shhhh

For some time now my podcasts have benefited from a semi-automatic transcript produced with the help of Sonix.ai. It is a paid service, but not all that expensive, and has the great advantage that it is very easy to correct the output because it has a nifty built-in player that will let me listen to the audio corresponding to weird words. Having corrected the transcript, I generally download it and gussy up the formatting a bit before sharing it as a PDF with each episode.

The PDF file is great for anyone who wants a reasonably accurate written version of the podcast, but it isn’t all that useful to someone who wants assistance while actually listening. Just such a person asked if I couldn’t also offer more of a verbatim transcript that can be read in some podcast players. Why not?

Whisper seems to be the go-to solution for this kind of thing, so I sent an episode to an online version and downloaded the result. As a TXT file it seemed pretty broken, just a single stream of not very accurate words. There was a different format with timecodes to indicate when the words were being said, but it wasn’t much use to my guinea pig. After a bit more research, I discovered that Whisper can supply SRT, a standard subtitle file, so this morning I spent a happy hour installing it on my computer and getting it to work on the most recent episode.

After a couple of glitches, easily overcome with the help of ChatGPT, I was astonished to find that it would transcribe a 23-minute audio file in about 50 seconds. To me, that is nothing short of miraculous. OK, so it mangled almost all of the proper names and a few ordinary words, but it didn’t take that long to zip through the file and fix the more glaring errors. That file is now associated with the latest episode and I am waiting to hear whether it does the needful for the listener who provoked this upgrade.

In the meantime, I took another look at Sonix.ai, and of course it too can give me a standard subtitle file. So here’s my new plan. Let Sonix do the transcription, and use its tools to correct it there. Then download TXT and SRT files so that I can continue to offer a readable PDF with the added bonus of the subtitle file.

We’ll see how that goes, and as ever, I am grateful to the podcast’s supporters for helping me to afford to use Sonix.ai.