Evaluating the Accuracy of Transcription Services: A Detailed Comparison
Dave Jackson tests transcription services by comparing their output to the original text. He finds a 30% error rate, questioning their efficiency and reliability.
File
Transcription Services How Accurate Are They
Added on 09/08/2024
Speakers
add Add new speaker

Speaker 1: Hey, this is Dave Jackson, and I wanted to take a second to look at, there are more and more transcription services, and I wanted to see exactly how accurate they are. So this is actually, this is what was said on the podcast. This is actually out of a Word document. I'm actually reading this particular document as a podcast, as a test. And so I went over, had the service, transcribe it, and this is what I got. It's a text file. Now, what I've done is I've selected this because this is me saying, welcome to the history of the band Six Shooter. And you can see it already missed the first word, welcome. And so just to make it a more apples to apples comparison, I'm going to delete that because that is not in the original version. And now I want to compare. I'm going to save this as a Word document, which I've already done, and I'm going to let Word compare these side by side. And just so these are kind of on the same page, you can see things like down here. I made a sound, you know, so that's going to be hard to match up, but I'm actually taking the heading here. This is the original document. This is the one I read on the podcast. And so what I did was I took my heading and I went over here and put the heading here. So it's starting off exactly the same way as the other one. And this is the transcript in this case from PIPA. And so now here I am in Word. I'm going to come up here and say I want to compare documents from two separate authors. One was me. The other one was from the transcription service. So I'm going to click on that. And I'm going to navigate here. Here's my actual text document that we just looked at. And over here, I'm going to say, give me the transcript. So here's the text transcript that I got. I took the text transcript, put it in here. And as you saw, I just added the heading. So they're both starting off at the same place. And I'm going to say, look, I don't really care about formatting. And I'm going to say, don't worry about headers and footers and end notes. I'm just really looking for in fields. Just look at the text, basically, is what I'm looking for. Case changes. Again, I'm just looking for the content. Comments, don't care about. I just check the text. Don't care about tables. Text boxes, nope, don't care about those. And move. Okay, so just check the text, please. All right, we're making it as easy, a fair fight as we can here. And when I click on okay. Okay. And you can see I have 424 revisions here. To go through and I'm just going to click this first one where I did something with the word and, and I'm just going to say, hey, let's accept that and go to the next one. And this is where things get fun. Yeah, now you can see all the changes. So if I were to, you know, it just, it goes on here. So six troubleshooting, it just so. There are services like Temi, T-E-M-I. I think that's 10 cents a minute. There are tons of these transcription services. And as the old saying goes, you get what you pay for. And so it may sound like a really cool thing, but realize, and this is why I don't use them. I can type faster than if I actually were to go through and fix all these changes. So for me, I'm not really gaining anything. I don't want to put this on a website for Google to find it. Holy cow. Google is going to say this is because then again, a human is going to read this. And I don't want a human to read something with this many mistakes or typos or whatever's going on. So just be careful when you see these transcription services, they sound really great. But at this point in June of 2018, they're not as cool as you think they are. You're going to have to spend a fair amount of time fixing all these changes. And so I went. I went back to my original document and I can see here where according to word, I have 1328 words and we had roughly, we'll bring it down a little bit. We'll round it down 400 revisions in that. So if I take 400 revisions and divide it by 1328 words, that's 30% error or 70% accurate. And. Use that to make your decision on whether or not these tools are worth the time and effort that it takes to clean them up.

ai AI Insights
Summary

Generate a brief summary highlighting the main points of the transcript.

Generate
Title

Generate a concise and relevant title for the transcript based on the main themes and content discussed.

Generate
Keywords

Identify and highlight the key words or phrases most relevant to the content of the transcript.

Generate
Enter your query
Sentiments

Analyze the emotional tone of the transcript to determine whether the sentiment is positive, negative, or neutral.

Generate
Quizzes

Create interactive quizzes based on the content of the transcript to test comprehension or engage users.

Generate
{{ secondsToHumanTime(time) }}
Back
Forward
{{ Math.round(speed * 100) / 100 }}x
{{ secondsToHumanTime(duration) }}
close
New speaker
Add speaker
close
Edit speaker
Save changes
close
Share Transcript