This April marked the beginning of the fourth year for Cuke Audio Podcasts. These days I'm only putting up one a week, almost always with a guest. Quite often I wish to refer to one of them but that's too much trouble with audio still. To search quickly, we need transcripts. I've transcribed many many audio tapes and files but I wasn't going to start transcribing podcasts. I knew there were voice recognition programs that did that and I knew they'd become more and more accurate. So I did some research and tried transcribing a podcast I wanted to get to with Mac Whisper. I got the pro version for $17 and opted for the biggest version I could get, big here meaning bigger file, slower, but does a better job. It did a good job fairly quickly. I'm not sure it's the best because I don't see how it learns. It might not learn - like iPhones and Dragon Naturally Speaking type voice recognition learn our voice. I ran the transcript through Word spell check and got through 20 pages in half an hour. Then I spent a couple of hours getting the whole thing somewhat presentable. One problem is that all the podcasts up to now I've mixed down to one mono file so I had to separate the guest and my lines. From now on I'll save them stereo, the way they are recorded, with the host on one track and the guest on another. Mac Whisper can then be told a name for each and the transcript will reflect that.
So here's the first one, still a bit ragged but good enough.
Yehudah Alan Winter - with subsequent emails
