SPEAKER_05 0:00

Explore AI. It's okay to give everything a try. And you know, using AI voices, using high quality can really increase the engagement and the professionalism of your videos. And if you want more learner retention in terms of the topics that you're talking about, the audio is quite important. So spend the extra time or a little bit of extra budget on that in order to kind of get the best video and the most professional video you can.

SPEAKER_04 0:28

Good morning, good evening, good afternoon, wherever you're watching from. And glad that you're here today with the Visual Lounge. We're I'm so excited about today because we're talking about some new research that we've just are about to release, or what by the time you see this, it will be released. And it's very exciting about using AI voices. And one of the things that's true about AI voices is I know there's a lot of questions, right? Like should I use AI voices in my instructional design, my training, my marketing? What we don't have all the answers, but we have some interesting insights that we want to share. And I am so excited to share them because it got me thinking differently about how I might use AI voices, when I might use them, and just really apply things. So I want to introduce our guest today. I want to welcome to the Visual Lounge, Stephanie Warnhoff. Stephanie is a senior marketing researcher for Tech Smith. Stephanie, welcome. Hello, thanks, Matt. Stephanie, why don't you just take take a second and tell us a little bit about yourself?

SPEAKER_05 1:20

Sure. So I am a senior market researcher, as Matt said, and I have worked at TechSmith for almost 16 years now. Um it's kind of surprising, but I actually went to school to study market research. So what I do here at TechSmith is I basically help our business make data-driven decisions. So instead of people sitting in a room and deciding where to go with Snaggit, where to go with Camtasia, how much to charge, what features to put in, we actually ask our customers, our users, people who could be our potential customers those questions. So I do that through mainly surveys, sometimes interviews, focus groups, online communities, basically all different ways when we can contact people to kind of make sure that their opinions are heard and that we go the right direction with what we're doing as a company. So if you've ever received a survey from us, it was probably 90% chance that it was written and the email was sent by me.

SPEAKER_04 2:09

So yeah. And one of the things that Stephanie, I think, will point out is if you've ever heard me talk about like the video viewer study and that research about how long should videos be, uh, and what types of attributes a video make for a good video from a viewer preferences standpoint, you have helped with all of that research as well.

SPEAKER_05 2:27

I have, yep, yeah, that's great. We've done that several times now. We'll probably be on track to do that again, either late this year or early next year. So that's exciting. It's always fun to refresh that. And I'll be interested to kind of incorporate some new um AI elements into that, which we haven't really touched on in the past, especially with the explosion of that. That will be certainly something that we want to find out about. You know, with as Matt's lead-in here with the AI voice and AI Avatar study that we recently did at the end of last year.

SPEAKER_04 2:53

Yeah. So Stephanie, let's let's dive into this new research. Uh, tell us a little uh give us the context in the setup for what we're about to learn about.

SPEAKER_05 3:03

Sure. So in this study, we created a very short 45-ish second how-to video on using Google Advanced Search. So our research team kind of tried to come up with a concept for a screen and voice video where we could kind of explain something quickly, but not something so simple that everyone already knew how to do, like how to Google something, you know, just straight up is pretty easy. People already know how to do that. But we wanted to kind of teach them a little bit more in-depth about it. So we came up with using Google Advanced Search. So there's one, like I said, 45-second video, and the screen content is exactly the same, but the only thing that changed is the voice quality. So we had four different versions of this video. One is a human voice that is kind of poor or maybe just not crisp and clear, right? There's some background crackling, and you can tell the speaker is farther from the microphone, so it just doesn't sound that great. Then there is a very professional level human audio, obviously a pretty nice microphone, um, absolutely no sounds in the background. Then the third is an AI voice, which you might notice is an AI voice. It sounds a little robotic and stiff, you know, there's no breathing, there's no pauses. It's just it sounds like a robot. Um, and then the fourth one is a what I would call a professional or high quality level AI voice. And it is um great. Uh, frankly, it is tough to distinguish whether it's AI or not. It is not robotic sounding, the emphasis is in the correct place. There are even breaths you can hear in the voice. So we have this video with four different versions of the voice and asked about 11 questions of each participant. We had about 800 participants in the survey. So we got quite quite a bit of data on this.

SPEAKER_04 4:47

Okay. Well, since you just explained those voices, should we go ahead and we we actually have the sample samples of those? Should we go ahead and listen to those now?

SPEAKER_05 4:55

I think that would help give a lot of context, yeah.

SPEAKER_04 4:57

All right, so here we go.

SPEAKER_03 4:58

Google search results can be a bit much. A simple question like what is the best sunscreen? Is full of ads and profit-driven biases?

SPEAKER_02 5:07

Google's advanced search resolves this. To begin, type in your search as you normally would. Go over to the tools menu and choose advanced search.

SPEAKER_01 5:17

Your search query will show up at the top. You can add additional terms such as toddlers. Scroll down the page to domain. It is the only research findings from universities. Type in dot edu.

SPEAKER_00 5:32

Finally, I click advanced search. When reviewing the search results, you might find more trustworthy answers from, say, the Cancer Center at the University of Utah than from Amazon.

SPEAKER_04 5:45

Okay, so you can you can really hear the difference between those different the different qualities. The AI the first AI one, which was the third one, really stood out to me. The the first audio, the human audio is actually I can tell it's background noise, white noise. It's not not great. So I a as we you started looking at the data here, I feel like there's always uh in research, there's always something that you're like, you got all your hypotheses, the things that you're thinking about, like, oh, this is gonna be um I'm curious, out of the gate, uh what was something that just like really surprised you in the study about the AI voices? Maybe there's something that was outside of what you hypothesized or what you expected. What what was maybe one of the big, oh gosh, here we go?

SPEAKER_05 6:28

Yeah, that's a great question. So I think just in being so close to this, designing the study, watching the videos, listening to the audio over and over again before we even launched it, in my mind, I thought, well, it's pretty easy to tell the difference between these. You can tell the difference between the poor human, you know, poor quality human and the high quality human. It's pretty easy to tell the difference between the poor quality AI and the high quality AI. But in actuality, it wasn't really that easy for people to tell the difference. There was a question later on on the ask them to identify whether their voice calls created with AI or not. And I thought, well, there was a question later on the ones that are created with AI to identify whether the voice created. That's obvious AI was not so obvious. And I thought, um well, the ones that are created with AI and the high quality AI voice, it's very difficult for people to distinguish whether it was an AI. Once you got to kind of the both actually quality, almost 50% for people to distinguish. I cannot tell if this is AI or not. And actually, which to me almost 50% or just because it just straight up like an AI voice. I cannot tell you. Probably also because I know the human being is a very good idea. Which to me, human audio and hear his voice often, it's probably like an AI voice. Probably also because I know the human that recorded that. Definitely voice because he recorded that.

SPEAKER_04 7:47

I guess it's just thinking about that, right? Because if people were uncertain, half the people were uncertain, right? Uh about some of those. Uh it it sounds like that, and and I want to be careful here, because research never tells uh like I I put caveats on research all the time. Like, research doesn't tell us what to do here. It just tells us it it's informative, right? Um, but it it does seem to see say uh you can start to think uh hey, with this fact that people can't tell, there's either a need for disclosure that I should be informing people that this is maybe not a human, or uh it gives me a little bit of room to say, like, actually I can feel more comfortable if I'm maybe uncomfortable with AI because people aren't gonna notice the difference. It gives me some room to feel maybe a little bit better about that. So does that does that seem fair from your kind of as you look at the all the data, or or should we be more careful here about like making assumptions?

SPEAKER_05 8:44

I do think that makes sense. And you know, we talk about AI obviously a lot at work and you know, things that and in even my career professional. I do think that makes sense. And you know, we talk about JI obviously on AI for which ones aren't, you know, is it not ready to replace it? And in even my career profession, about you know, what parts of your job can be. So we talk a lot about AI as a tool, you know, is it not ready to replace a human? You know, just you know, outsourcing everything. So we talk a lot about AI as a tool in your system that you should not probably be you know outsourcing everything in your system, which it is, not a human's voice. It is a computer, computerized voice, whether it sounds like a human or not. You know, fake. Which it is, it's not a human's voice. And I do think computerized voice, whether it sounds like a human or not, it is important to people. But it is a tool in a tool, right? They maybe couldn't distinguish in the survey whether those high quality voices were artificial intelligence or know whether it's AI. They maybe couldn't distinguish in the survey whether it being true. Those high quality voices were artificial intelligence or pose something or people do want to know and pass off something as AI and you don't disclose that pose something erodes the trust in your brand. Pass off something as AI if you're using it and you don't disclose that. I think it erodes the trust in your brand. It is important to disclose that if you're using it as a users do want to see that, especially we saw some differences in this survey. We did ask about you know disclosure, and there were some differences. You know, our users do want to see that especially we saw some differences. In this survey, we did ask about wanting that disclosure than people in Germany.

SPEAKER_04 10:18

Um respondents, I believe, that were uh that's it came up as I was looking through this people in North America. A lot of the people for the AI voices uh rated the voices as like high, like professional, right? High quality. Um and I'm curious, um, even and that even matched like the human quality, like it's professional on either side, high quality AI voice, high quality human voice. Um based on that kind of looking at like people maybe didn't uh always be able to put pull out whether it's AI or not, plus people thought it was really high quality or professional seeming, which I think is important for good good instruction, good, good kind of whatever video you're doing. Um Are there any specific markers that like maybe signal high quality or professional voices to to maybe learners or or viewers of video today? And I know that's not in the study, so I'm putting you on the maybe on the spot a little.

SPEAKER_05 11:10

Yeah, that's okay. Yeah, we didn't really focus on the professionalism aspects within the video, just whether, as a general rule, this video. Yeah, that's okay. Yeah, we didn't really focus on the professionalism aspects in the video, just whether as a general product. This whole video was professionally produced. You mentioned, and as I have running 92% of people felt like that quality of the surveys of the case. This whole video is professional for 15 years, but as you mentioned, go back to the video. I have run in this example a lot of video view studies. Um, one of those things, go back to the listeners know that that is the video unview study that we did in 2020. Early professionalism and in that study, I think they're pretty tightly tied a video is more unviewable and there are aspects of early professionalism. Two things are pretty tight to video and keep their attention. You know, when a video is professional and retain the information that they saw in that video. And so just thinking about what characteristics are more likely to retain the information that they saw in that video. You as a viewer, we didn't ask what characteristics of a video. What characteristics you as a viewer engaged in a video quality keep them engaged in the video quality? The difference between that very nice and high quality can be sometimes that audio quality with that background noise or with that mic that is a very difficult time. The difference between a lower quality and using your laptop, can be sometimes challenging or background noise hard with that mic that is either you know a really quality microphone, you know, using your laptop sometimes can be challenging or being a video hard to see. So that is like the number one thing. There were some other things in class study, also like professional style. Um professional looking graphics like text. So that is like the number one third, you know, like there were some other things in mass study. I mean, other professional style eyes on the screen, like mouse cursor text overlays or lower thirds you know like in this AI. Um and other things like voice study their eyes on the screen. Like mouse. I would like to kind of test some of those things, which we didn't know later on to check and see if those things lead you and voice study. People thinking um you know perceived professionals. I would like to kind of test some of those things, you know, later on to check and see if those things lead to the other.

SPEAKER_04 13:09

One thing that I think is really interesting to me about these things. A lot of the work that I've done and people that I think listen to the Visual Lounge are in the world of learning, right? They're they're helping corporate learning. Um, and I I can't think of any of our other studies where at least not I've I've seen the research that you've done come out of this that we're putting people in kind of the situation where like, you're gonna watch this, and then there's this pop quiz, right? Like, hey, did you actually learn and retain? Which is obviously a big deal if people are able to learn and retain from one style over the other. Uh, was there something specifically you're trying to isolate in the study about learning or retention of kind of knowledge in what you did?

SPEAKER_05 13:49

That was an interesting one for us too. I kind of call we sometimes we call those behind the scenes like uh put your money where your mouth is question, right? Because we're showing you a bit of information that excuse me, that was an interesting one for us too. I kind of call we sometimes call those personally amnotal, put your money where your mouth is. That was a new topic for me. We're showing you a bit of information that a long time ago. You know, I that's personally amnotical, but user of Google Advanced Search that was able to learn something. Like I mentioned, so we did debate. Show where to go in Google Advanced search during the video. It was spoken a lot. Go here, um, if you couldn't learn videos 45 seconds. So we did there were about 11 where to go in Google questions was maybe in the video of seven. It was more towards here. And there was no videos 45 seconds. There were about 11 questions. We'll go back and do that question was maybe numbers. Counting on the phone. It was more towards paid attention to that video. And there's no ability to go retain that information all the readings. Or go back and do that video. So we were counting on the phone. So it was multiple changes. Paid attention to the ability to find the beginning search, retained that information through all of those. Correct what they heard. So it wasn't what they choose, we said to go to find groups. And as it turns out, then there were four or five questions. For the most part, what they heard. Especially in the video. We have a higher percentage of correctly. The video with the high quality AI for the most part, AI actually. Especially in the AI side in that pop quiz question. Higher percentage of things that kind of surprised me. I kind of saw it, you know, the high quality will be the low quality. AI actually, but that wasn't true. It was the AI in that pop quiz question, which means that's another thing that kind of surprised me. That kind of high qualities will be the low qualities as a pair. But that wasn't true. It was the AI people were paying more attention to that. And did so that's just another plus point for including AI if you have to.

SPEAKER_04 15:38

I'm gonna ask it, and I think this is gonna be one of those things people are always gonna ask when they look at the data. What the research also, what the data showed as well is that the professional sounding human did slightly less well than the poor AI or the poor human quality, which is which is you know, like we just talked about the importance of good quality of voices uh and audio, right? So that was surprising to me. It's and it's not a big, it wasn't a big, it was within a margin of error, I would, I would assume, right? Just a couple points difference. But that really surprised me that the human professional quality sounded sounding voice videos did a slightly less well than the the poorer quality human videos.

SPEAKER_05 16:20

Yeah, and I don't I don't have a direct answer for that because we don't know exactly why, but after looking at this data and kind of reviewing those videos and some of the open-ended comments, we kind of have a hypothesis. And I don't I don't have a direct answer for that because we don't know exactly why, but after looking at data and kind of reviewing those videos and some of the things that we're gonna do. People may have been straining purpose at least as closer attention because the words were actually harder here than in the human higher quality examples. You know, further away, people may have been straining in that they were paying closer attention because the words were helping them retain that information higher quality examples to retain that information a little further along. They were paying closer attention correcting. And that correctly I didn't retain that information. I would not recommend retain that information a little further along the journey. If you correct terribly correctly identifying, that's kind of our hypothesis. Now I would not recommend the margin of error, but it's audio exactly matched and think that everyone is gonna get 100% if many of it terribly. But that was kind of our hypothesis. Well, I think it's always the margin of error, but it's not a good thing. Exactly.

SPEAKER_04 17:24

One research study is never enough. That's why you gotta keep doing things and looking at things. Um and and speaking of research study, there's there's actually been a lot of uh various research over the year, and uh obviously AI voices have come a long way in the last year, two years, five years, right? Um, and there has been suggestions about like hey, you need a human-like social presence for deep learning. That's uh Richard Mayer, who does research, but uh more some more recent uh research, uh Nazimi Dinsier, I hope I'm saying that right, suggested that maybe there's not as big a as a difference in cognitive load when you've got modern AI voices. Obviously, when it sounds human, you think it's human kind of thing. Um Do you think there's an effective like uh anything in the research that's like about effectiveness of learning and retention that with AI that we we haven't talked about already? Is there something else we should be considering? If I'm a practitioner out there and I'm looking at like, hey, do I want to use these high quality AI voices, whether through TechSmith products like what Audi offers from 11 Labs or some other tool that's offering high quality voices, uh anything else that we could we can do to boot bolster kind of retention or understanding from those? Or it's a lie packed into that statement.

SPEAKER_05 18:39

Yeah. I think that um based on this research, if you're thinking about trying it, you should. Um that's basically the recommendation I would give from this, especially if it's a high quality AI voice. I think that um based on this research, and when we're thinking about trying it, our research certainly shows basically the very high quality AI voice, especially if it's a high quality AI voice. If you as a matter of fact, looking at that pop quiz question, the number of research certainly shows the percentage of people that have high quality AI questions most correct were people who have the high question. As a matter of fact, looking at that pop quiz question the end of the scale, the top of the percentage of people that had got that question most correct. If you have tried quality, it's a good recommendation. They had content creators, the highest percentage that you can measure that. If you're thinking about trying it, then it's a good recommendation you would try. But is there something that is worth trying and it's certainly not gonna turn off your audience when you unless you know about quality? That's a difference. But Rachel is AI something that is worth trying, and it's certainly not gonna turn off your audience unless you're talking about quality, then that's what I'm saying.

SPEAKER_04 19:48

Research does for me as a practice. AI as a blanket statement is designed practitioner work, is that it allows me that room to say, it's okay to try this, we're not gonna ruin things. But obviously, every environment, every circumstance is a little bit different, and you gotta the things that we were saying on the visual lounge all the time, know your audience, think about the outcomes you're going for, all those ch all those types of things still are in play. Uh, but it does give an encouragement that like this could be really effective, and particularly for certain types of videos where it feels like there's more opportunity. That this makes sense, right? Things are changing all the time. You've got a video that's uh instructional video that's a process that's gonna update three times a year. Like an AI voice makes a lot more sense than maybe having someone have to go back into the studio to record. I I am curious, Stephanie, as you as we're getting close to wrapping up here, um you talked a little bit about kind of what your hopes are for uh future research. I'm I I want to go there again. I do want to note to our listeners, if you're still listening with this, uh we're gonna meet again. We're gonna Stephanie and I are gonna talk again about uh the second half of the study, which was AI avatars and what we found using the same process, but some different styles of avatars and approaches versus humans. Um but is there things that you if you could, you know, magic wand do uh the more research in this area about AI voices, anything else that you would really love to be able to dig into?

SPEAKER_05 21:14

Yeah, that's a great question because oftentimes when we complete a research project, we have more questions than we started with because it is inspirational. That's a great question. Well, if we tried this, or what if we did that project? So we oftentimes end up with have more questions than we started on the one. It is inspirational to kind of information and say, well, what if we tried this, or what if we're about to do that? So we oftentimes end up with um another question with some markets and engagement. There are somebody around with this. We're not learning in um related to changing markers and size of the mouse. I would like that kind of to play around a little bit more necessarily in AI, but it could be you know that standard related to 2x or maybe 10x of the mouse. That leads to higher professionalism or more engagement video. Standard size or 2x or maybe 10x where the video seen around professionals or something, some of the video enhancements that we can change, keep them. We can help engage them, but also not better. Distract from the actual functioning and what you're showing on screen overall. Some aspects in the video. I'm hopeful that we'll be able to change and engage them, but also not distract our kind of actual learning and what you're showing on screen overall. So be really good. I'm hopeful that we'll be able to do that sometime. If you do any of that research, let me know. Because I talk about this stuff all the time.

SPEAKER_04 22:36

And I've got hypotheses about some of those things based on what I've read. I love that TechSmith is willing to invest in these things because I I benefit as a creator oftentimes from the research that we do and helps me uh think better about our our projects and the videos that we create. Um as we we're about to wrap up, but anything in this study about the AI voices that we didn't cover that you feel like, gosh, people need to know about this. They might be in the report. Maybe it's something that didn't make the cut for the report.

SPEAKER_05 23:08

Yeah, I talked about this very briefly at the beginning, but the thing that was probably most interesting to me was the uh indistinguishability of the high quality. Yeah, I talked about this very briefly at the beginning, but the thing that tells me most interesting to me was the it's okay to try distinguishability as the high quality versus a strategy as a tool that's tells me and we talked about try one video with a high quality AI this text uses your own as a part or your own creation strategy. Just to see try one video with a high quality AI if your viewers are more engaged. Your own recording or your own retail voiceover expert tendency. It's okay to do a little trial and error. And I think what this study tells us is that viewers are more engaged and retained in terms of things about it. It's okay to do a little trial and error. And I think what the study tells us is some are positive, some are negative. But if AI is not allowed, you can get to hear things about AIs in all aspects of a high-quality voice. But some are positive, some are negative by the case. But in this situation, you can give it a try. Necessarily turned off by anything. There are a lot of positive comments and go back to open human voice here. It's certainly about to hinder what you're doing. AI in general, and how aren't the voice sounded necessarily turned off? There are a lot of positive comments in human as long as I'm getting the information I need, give it to me in the most pleasing way in general. And how do you think that's a good idea? So I think it's worth a shot and didn't necessarily care that it was a human.

SPEAKER_04 24:33

As long as I'm getting the information I need, give it to me in the most pleasing way for my own. Well, Stephanie, this has been fantastic. I'm I'm really looking forward to our conversation about avatars because I know there's lots more. I think AI voices are is is a little bit more tame. People are a little bit more accepting. AI voices will be a little bit, we'll get into some good stuff there. Um I I do want to ask, you know, I know we do a lot of research studies at TechSmith. If people wanted to be involved in something like that, or if maybe they want to reach out to you, whatever you feel comfortable with, where what would you recommend they how did can they get a hold of you?

SPEAKER_05 25:11

Yeah, that's a great question. I have LinkedIn. If you'd like to personally connect with me on LinkedIn, it's Stephanie Wernhoff. But if not, you can email. We have a research at TechSmith.com. That's a great question. I have LinkedIn if you'd like to personally connect with me on LinkedIn to our panel. But if not, we can't email we have a in order to get you selected. You know, future get you added to our panel. But more about you. Interview projects and selective scores and different future type things all mixed. Honestly, we send probably that is the best way. Surveys every month all year long and with interview projects and focus groups and different community type things all mixed in.

SPEAKER_04 25:49

Perfect. And I will mention that is the best way to do that. I don't think we said it. Maybe we did, but it's worth reiterating email. Everyone in the research study that we just talked about, they were not customers of TechSmith, correct?

SPEAKER_05 26:00

Correct. That is correct. They were what we call non-customers. So we use uh a provider, a panel provider, there are hundreds of these online to find people that are not currently using our tools. Correct, that is correct. They were what we call non-customers. So we use study a provider, a panel provider, there are hundreds of these online to find people that are not currently using our tools. So it's not study of text with customers. Give us a quick wrap-up who basically at least once a month. Um my final take would just be that um, you know, explore AI. It's okay to give everything a try. And um, you know, using AI voice. Um my final take would just be that increase the engagement. Explore AI to make everything a try. Increase the engagement and the professionalism of your spend the extra time or a little bit of extra budget on learner retention or in terms of the topics that you're talking about. The audio is quite important.

SPEAKER_04 26:56

Well, Stephanie, thank you for the same.

SPEAKER_05 26:57

Spend the extra time or a little bit of extra budget on the um in order to kind of do it.

SPEAKER_04 27:01

All right, so some really great stuff, and go check out the research report, we'll link to it down below. But there's so many great things about audio, and I and I love the this approach that like it opens up this opportunity to really start to digging in and not feeling like, oh gosh, AI audio uh is going to ruin my project or it's not gonna be uh good enough. It's not professional. Uh and if you're looking for an AI voice provider, you might check out Audi. If you're buying the individual license, we now have 11 labs, it's fantastic, some amazing voices in there. I'm always impressed and blown away when I go and choose one. I'm like, wow, that's really good. Uh and some other great options as well from so lots of options in Audi if you're looking for one, or there's lots of other great providers out there. But uh with that said, you know, think about what you're processing, think about what you're doing. Obviously, AI is infusing itself into everything we do, so we gotta keep talking about it. But but we want you to make better videos and we want you to make videos that are effective for the job that you're hiring them to do. And we're so grateful for you listening, and we hope you take a little time to level up every single day. Thanks everybody.