There’s no I in my AI

A few years ago, I was having a phone interview that a friend of mine setup. I really wasn’t looking for a job. But this was a well known company, an industry standard, and one whose products I’ve used and liked for years. It went well enough, however, my friends boss told him it didn’t seem like I was interested in the job.

No shit. I wasn’t, and I told her so.

But one of the questions I was asked was what do you like about you’re current job? What do you not like?

Oooh…good questions. I’mma use them next time I interview someone.

The best part of my job at the time was getting to play with cool software and explain it to people, as well as getting out of the office go among the living to do pro services work.

The worst was that point during the elearning development process where I’d spend days, if not weeks, prepping content to actually build the course. This consisted of recording and editing audio, collecting and creating graphics, researching tech manuals to make sure I get stuff right. Turns out, that was 100% of the job she was looking to fill.

And that’s where I find myself lately.

I’ve been given the task to update some courseware that was created by a guy who retired. There’s no script. So I have to send the audio through Dragon, because I have a copy, have it turned into text. Fix it, update it, and then I have to run it through Speechelo. The reason for this is so what happened does not happen again. Although it would be way faster for me simply to re-narrate it.

So I feed a chunk of audio into Dragon. This is what came out:

The first step. Courses afflict the chart type no. When you Dragon chart work will be prompted to choose chart type nurse Alyssa we have available to us here are lying. Pie meter scanner start bubble difference get to and coat colors and variations of our notice under the subtype, you can select how you want the data to be displayed in the chart…

Hamburger from Dragon

Here’s what it actually said:

The first step, of course, is to select the chart type. Now, when you drag a chart into a report, you will be prompted to choose the chart type. Notice the list we have available to us here; Bar, line, area, pie, meter, scatter, stock, bubble difference, gantt, tube and cone. Now tube and cone are just variations of a bar chart.

Notice under the subtype, you can select how you want the data to be displayed in the chart…


To be fair to Dragon, you generally train the speech engine to your voice and speaking style. Since this is not my voice, I’m using a generic transcription profile. And,

  • The audio quality is poor
  • Dude has an accent and style that I’m thinking is Pennsylvania. Definately a mid-atlantic accent and style.
  • His cadence is for shit. He speeds up phrases, and then pauses.
  • And most of his sentances are not complete, his grammar and diction aren’t the best.

That’s a tough nut for a speech engine. I’ll have to do substantial rewrites to make sure it doesn’t sound retarded in Speechelo.

The thing here is ‘Garbage in, Garbage out’. I’ll have to do a ton of prep so the AI engine of Speechelo can make sense of it. Which is true of a lot of AI. They really don’t learn. They simply add to their database and look stuff up faster.

It’s a lot of work that causes me to sit at my desk for hours at a time.

Have to take a lot of breaks. Do small chores, take a walk, play with the dogs.

To quote Bugs Bunny…”eh…it’s a livin'”