You Can Always Catch Me On Sweater

This morning I was catching up on messages after taking the day off yesterday.

I’m on the planning team for a higher education conference this month, and we have some details to finalize in the last few weeks before the event.

I needed to follow up with the lead conference organizer, and left him a voice message about registration.

Here’s what the lead conference organizer read in his email:

Hi Amanda it’s Deborah Edwards Miller oh I was out yesterday so I am catching up finally email that went through I wanted to confirm your registration clothes.

There are three tweak or but to treat some people asking specifically about registration and there’s been no response

— so I’m trying to get some answers to the or wondering if I may and contacted through direct message so there’s no indication — that they have been responded to so I wanted to double check.

Shannon let me know (xxx) xxx-xxxx you can always catch me on sweater thanks.

Oops!

I have to say I had a good laugh when he replied back, with the text of my voice message.

But, what in the world?

That’s not the message I left. The lead organizer’s name is Matt, not Amanda or Shannon.

And you can always catch me on sweater? Where did that come from?

Like many people who work at large organizations, the lead organizer uses a voice mail system that integrates with email, using automatic speech recognition (ASR).

He receives a text version of the audio recording he hears when he listens to his voice mail. The text version is included in the email message he views in his email application.

I thought I spoke slowly and clearly when I left the voice message, but speech-to-text conversion varies in accuracy. And that’s what happened with my message.

So, what message did I leave for the lead organizer? Take a moment to listen to the audio.

Or read the message I left.

Hi Matt. It’s Deborah Edwards-Onoro. I was out yesterday, so I’m catching up on all the email that went through. I wanted to confirm: is registration closed?

There are three tweets, or is it two tweets, from people asking specifically about registration. And there’s been no response.

So I’m trying to get some answers to them, or wondering if Amanda contacted them through direct message. There’s no indication they’ve been responded to. I wanted to double-check.

Can you let me know? XXX-XXX-XXXX or you can always catch me on Twitter. Thanks.

Makes more sense, doesn’t it?

The automatic speech recognition converted many words correctly, but missed quite a few. It didn’t understand where sentences ended. And misunderstood names and verbs throughout my message.

And yes, I told Matt he could always catch me on Twitter, not sweater.

Challenges of Automatic Speech Recognition

If you’ve ever used YouTube’s automatic captions, you know automatic speech recognition isn’t 100% accurate. It’s dependent on a number of factors including pronunciation, accent, pitch, volume, speed, echoes, and background noise.

Whenever anyone asks me about using automatic speech recognition for podcasts or YouTube automatic captions for videos, my reply is that it’s a good way to get a first draft. And that you’ll want to edit that first draft to ensure the conversion is accurate.

Updated: Automatic speech recognition continues to get better. However, if you’re using speech recognition software for transcripts, having a person edit the automated results will always produce a better end result.

How have you used automatic speech recognition? What have your results been? Share your experience in the comments.

And don’t forget, you can always catch me on sweater!