Pulse Adds Audio Description Audits for Title II Readiness (Full Transcript)

[00:00:00] Speaker 1: Welcome everyone. Thank you today for joining today's session. Titled new from Pulse audio description built for title 2 compliance. My name is Jacqueline. I use she her pronouns and I'll be moderating today's webinar. And with all that taken care of, I'd like to give a warm welcome to today's speaker, Eric Ducker. And I'll pass it off to Eric to share what's new with Pulse.

[00:00:26] Speaker 2: Awesome. Thank you. It is quite warm today here in Minnesota. We have the heat wave. I'm sure anyone in the Midwest can empathize. As you guys may or may not know, I'm I'm Eric Ducker. I lead product marketing here at 3Play Media. I use he him pronouns. Excited to do a quick preview of of what we're focused on with audio description and our product suite Pulse. I'm going to talk a little bit about, you know, just reminding everyone of the audience what Pulse is all about at 3Play, how it differs from what you might be familiar with with with 3Play services. And we're going to do mostly a live demo of the service so that people can get a preview of what's happening. And then ultimately want to talk about how we can get you guys engaged in testing out the service, using the service and seeing seeing what feedback you guys can provide to us. So taking a step back on kind of what Pulse is and Pulse is basically a solution that we announced at the beginning of the year from 3Play Media. And it really comes down to solving kind of three key aspects. One, what's the what's the problem, the visibility problem that we experience in compliance, how it impacts our budget reality and ultimately the urgency of of leveraging a solution like this for Title 2 compliance deadlines coming up in April. But starting with the visibility problem, one of the things that we have learned from, you know, interacting with customers over the last two years discussing their Title 2 plans is there are great accessibility checkers out there. There's compliance tools that help administrators or faculty understand the relative accessibility of their courses and other mediums that they use. But largely, once we get to video, it's a bit of a black box of what's happening. The accessibility checker might see a specific caption track is there or not there, but it won't tell you how accurate the captions might be. So you don't know if you're getting AI captions or human corrected reviewed captions. You're really not exactly sure what level of accuracy those captions really represent. In addition, you don't really know whether or not this video needs audio description. Not technically speaking, not all video requires audio description. I'm going to do my best to make this video not require audio description. And when we are done with this video, we will run it through our own scoring system 480 recommendation and share it with you and share that exact report with you about how much audio description this video requires. So I'm going to do my best. I don't pretend to be the best at presenting in this format, but I'm going to do what I can. But the visibility problem is real, and we hear this all of the time. And with captions, it's a little bit of an easier problem, meaning we have a decision to produce captions or not produce captions, and it's usually we're going to produce captions. But with audio description, now budget becomes even more of a reality. Cost of audio description relative to captions can be nearly 10x the cost, no matter the system, no matter the process that you use for audio description. It can be quite more expensive. And so we're not here. I'm not sitting here in Minnesota telling you that you don't need to do audio description on certain videos. I'm telling you we have tools that help you prioritize which specific videos are going to have the most impact of being provided an audio description track relative to all the other videos that you might have and allowing you to make the most of your budget and make the strongest impact on your budget. This ultimately comes down to being able to hit the compliance deadline in a way that it becomes defensible, measurable, trackable, and you can make improvements over time. Couple things that are going to happen naturally. AI is getting better and better, so your accessibility programs naturally year over year are just going to get better without you doing or lifting a finger. But then there's also the question of, well, what if we had more budget? This is what we can accomplish. So those users, those customers who are at scale using Pulse are having very different budget conversations than those who don't have access to these tools. And they're able to defend and make a really, really concise budget request to their administration and share the tradeoffs of producing or investing in this or not. So that's at the high level what Pulse at 3Play Media does. And I really want to just focus our time today on audio description and the latest innovations that we've done with audio description. Captions, we have plenty of webinar content on that from previous material. You can learn more about that in some of our other webinars. And I'm sure Jacqueline or one of our moderators will drop links into the platform, into the chat box. So the Pulse workflow, solution overview of the Pulse workflow is relatively simple. You're going to first connect and link your video platform. You're going to then be able to audit which Pulse will analyze every video file and assign scores for the predicted accuracy of ASR or AI captions, automatic speech recognition. And then you get to automate or design a remediation workflow where maybe specific files below an accuracy threshold go straight to human remediation or certain files go to an administrative review before going to remediation. These are things that you can look at for within Pulse. And then finally, moderation at scale, continuous scoring on all new content. You can view real-time details on your Pulse dashboard, and we'll actually show that today in real time. So let's talk about why 3Play Media and how we think that we're doing something unique in audio description. And before we get there, we are a partner that knows higher ed very well. Our first customer was MIT back 18 years ago, serving closed captions at scale for an online education course. We have developed solutions specifically to solve the challenges of higher ed with our audio description service. We have introduced AI-scripted audio description services. We've introduced and we've built that off the expertise of our team of audio description experts. We have dozens of audio description experts across our business that are writing scripts for anything from Hollywood content that you see on HBO and down to education content or across to education content in the classroom. And the technology that we're showing you today is specifically the technology that we use every day to process and create the outputs of audio description that all of our customers have become reliant on. So none of this is net new technology that hasn't been tested at scale. It's truly technology that we are opening up back to our customer at the end of the day. So what is this audio description service? We're ultimately building a service that answers the question, do you need audio description? And it's effectively works as an audit model. And what we are doing is analyzing the video and exploring what kinds of visual information are present in the video. So on this slide, you'll see a series of texts, sorry, a series of boxes with different types, different categories of text and visual information that we are measuring as part of the video content. So slide deck text, background text, copyright, diagram text, interface text, scene setting text, speaker identification, URL text, maps, infographics, tables, physical demonstration, image labels, diagrams, timeline, charts and graph. And I appreciate our ASL interpreter for keeping up with that one. But ultimately, these are all different visual informations that are being conveyed in the video. And the question is, how do we measure what needs description? Fundamentally, this doesn't necessarily cover intent. This doesn't cover the art form of audio description, but really gives people a full heuristic overview of what is present in each video. So I'm going to swap over to the product itself. And so give me a second while I transition. I'm going to stop sharing. And I'm going to make sure I'm logged into 3Play before moving forward. And then I will share my screen. Okay. So everyone should be able to see my screen now. And what we're seeing is just a single video and the output of the slide that I just talked through. So what we're measuring are all the categories and the percentage of that category of text that's present in the video. That's not covered by the dialogue. So we don't necessarily care about content or text that's been described by a dialogue. So we only surface information about visual information that captures what's missing from the description. So in this example, down here in the category breakdown, we have a slide deck text with a percentage of 16.1%. So for the duration of this video, the model has output that 16.1% of the video contains slide deck type text that is not being covered by the dialogue. Doesn't mean that that's important text. It doesn't mean that it's the most important text or the least important text. It's just categorizing the text and allowing customers to better get more fidelity over what is actually happening in the video. So we do that for all of the other features of text and visual information. So diagram text at 5.2%, URL text at 0.2%, charts and graphs at 24.4%, and so on and so forth. You get the idea. We're measuring at very specific intervals. We also want to summarize this into a very basic score. So if you don't have the interest in seeing the breakdown at the category level, we also just assign a mathematical output of how from a 0 to 100, how much does this video potentially require audio description? 100 meaning definitely requires audio description, 0 meaning, hey, there's probably nothing that's going undescribed. So for example, a video that has black screen going for an hour and just dialogue over it like a podcast, that would have an output score of 0 on this scale, for example. So this is helpful for the overall video, but now, once again, this is at the video level. And so we want to better understand where in the video we might be missing description. So we've also, the way that we analyze the video is we actually have both a video player here on the left side of the screen mapped to a heat map of where description may or may not be, where description ultimately is needed. So this is a gradient driven heat map where we have tiles from white to dark magenta or bright magenta, whatever you want to call that, indicating where in the timeline of the video we're seeing the most opportunity for description from a missing perspective. So if I skip ahead to this part of the video, this will allow me to just review this segment of the video. And this segment might be 20 seconds or 40 seconds. The model is a little bit flexible in those chunks by design to make sure that we're capturing an analysis where there are adequate breaks between speech so we can segment the video appropriately. But this allows us to review, hey, what might be missing from this particular content creator's dialogue track that's not being covered by the audio description? So if you're doing this manually, you can just skip to these pink points, these magenta points in the video, and it will scrub to that point in the video. You can do a quick review. So instead of watching this entire, you know, 26-minute video, you can instead watch maybe just a few minutes of this video or identify, hey, I need to go do description on this because it is actually enough is missing from our policy, i.e., the customer, our policy as a customer, what we determine. So from there, you can choose to remediate this yourself. Of course, we give tools that are integrated into our marketplace for remediation. So you can order directly from here the different levels of service from 3Play Media, everything from AI-scripted audio description up to our accommodation level, which we have been delivering for many, many years, which is a human-scripted audio description. One thing that I like to call out is that we are mindful of the fact that extended and standard audio description are important nuances in terms of description. And so we do provide a 3Play recommendation of what description type you should do. Largely, we're going to find in academic content that extended audio description is appropriate. But we don't ban you from ordering standard. We'll just make sure that you know that there's risks in ordering standard audio description. So we really want to give you the control to really understand what might be happening when you order audio description and why you might see a certain type of output at the end of the day. So this is really helpful at the individual video level. I need to do this once or a dozen times. But now schools have thousands of videos that they need to look at. And so this is where it's really important to think about, OK, how do I automate this across several files? And so this particular account has 10 hours of content that's been processed through Pulse. And we get to see a few things. We can start exploring the data. So we see where videos are falling into the distribution. We're starting to see a pretty standard distribution of all the video. So this will allow us to better understand how much budget we need to apply to make sure that we're getting the highest risk files described. And then maybe if we can get over to the left here, which has these bar graphs of buckets of 0 to 25, 25 to 50, so on and so forth, we have the duration on the y-axis and the AD rec score on the x-axis. This will start allowing us to see that bell curve and see where in our budget might we want to maybe set automation up. We also support, we also give you the opportunity to explore the advanced data. So you can see the slide deck text distribution. So the same kind of principle of the graph, you see where slide deck text typically falls. And you might actually stop there and say, OK, 10 minutes of my content is missing a lot of slide deck context. So I might actually go here into the explore specific files. And I want to look at any file that exceeds, say, 25% of, actually, this is the wrong category, sorry. In the slide decks category, I want to see anything that exceeds 24%. And that way I can take a quick look. Maybe there's a common content creator that's failing to meet our expected standards of description or universal design description within a presentation. So you'll see a filter that allows you to go into a specific file, go back to that exact detail page that we have. These files, for example, were remediated. So they've already had a description service ordered on them. So you'll see that labeled. Going back to the data side, though, is this allows you to start kind of seeing patterns. So I go from slide deck text to, say, charts and graphs, another common thing. All right, clearly we're seeing a lot more undescribed charts and graphs. How can we maybe do an internal training with the faculty on best practices for charts and graphs audio description? So lots of second order effects that can come out from this data from how do you help improve your accessibility programs at your university. But you could, of course, use this data to set up specific remediation workflows. So we've introduced this Pulse Playground. So I've scrolled down in the page so that the Pulse Playground, similarly to everything else, has basically basic and advanced settings, which allow you to simulate and automate this process. So, for example, if we want any file that has an 80 score above 75 and any score above 75, you'll have a slider bar or an input here that allows you to send that directly to remediation automatically. And you can customize these how you see fit. And then, of course, you can simulate them. So on the right side, you can press simulate. Doesn't order anything. It just gives you an idea of what you might spend. How does your setting impact your historic data set? And maybe give you a prediction for future spend on the simulation. You can apply these settings and then future, apply these settings in these future calls, sorry, future files will then have that remediation rule or review rule set up. Of course, basic might not give you enough fidelity. You have, you know, this still doesn't help you understand, well, what text, what visual information is missing. And so it's important that we also provided advanced settings so that you can look at the specific categories and make your own determination of what matters to you. So instead of having us give you a basic score, maybe you care about slide deck text. So you can create a specific rule about what types of files get triggered for slide deck text remediation or review from an admin in this account. And then you can have these, you have multiple of these. These are all working as or statements from a logic perspective. So if slide deck text or charts and graph or diagrams trigger a remediation job, the whole file goes to remediation. If none of those files, none of those settings trigger remediation, they go to review. If none of those files trigger a review, they go to automation. So here are some settings. I simulated it. You see much less action required if we just focused on these three visual types in our graphs. Now the auto-remediated is 18 minutes versus I think it was much higher. It was 97 minutes when I redid the simulation. So a much lower need for audio description based off of the rules that the university sets. So I promise you I did not do a great audio description job in this particular interface. So I apologize if I did not meet the standard that we were hoping for. I'm hopeful that we'll see that illuminated in the results of the description that we run into. But this is basically the workflow. And I'm going to go back to our slides and wrap up as we come up against time soon. All right, let me try to get to there. Okay, so finally just to wrap up a couple things. 3Play Stance is not telling you, we're not telling you what requires audio description. Our goal is to give you the tools to measure what's essential for audio description. So we ultimately believe you, the customer, can control and have responsible audio description practices. We're giving you measurement. It's essential for the management of really optimized audio description coverage. And ultimately, we see this as a progress in the pursuit of perfection. If we assume that we have to do audio description on everything, we're never going to get that budget. If we can start getting the administration and the university used to doing some audio description in a very logical way, we're going to move much further down the line towards perfection. So this is really an opportunity for folks to really start having a real conversation with the administration without scaring them. And so one of the things that we believe strongly about what we've built here is we want you to try it. And we're not saying just like send us a file. We want you to send enough content for free that allows you to really start seeing those patterns. That account that I showed in this demo was just 10 hours of content, and you could already start seeing some patterns. So at 50 hours, we expect most customers to see all of the patterns that they can. They never have to talk to 3Play again after this. You never have to buy anything. It's more how can you have a real conversation with your administration without the data. At the end of the day, this is your pathway to getting the data, and we want you to try Pulse for Campus-Wide Compliance. We think you're going to love it, and we want your feedback. So 50 hours free audit, both the captions and the audio description, Pulse service will be included. We already have about 10 universities onboarding into this pilot. We want to add many, many, many more. So I do encourage you to sign up. It's really, I promise you, no strings attached. I'm not trying to be, you know, salesy here. It's just we think that there's so much value in this data, and helping you shift the story and the narrative within your university. Worst case scenario for you, you get a bunch of data, and you can start that conversation. Best case, you find out that this is solving a huge problem at your university, and you want to give us a ton of feedback so that we can build it the exact way that's going to work for you and how your content's created and monitored. So I'm going to pause and let Jacqueline come back with questions, because I obviously did not look at any questions coming in.

[00:24:27] Speaker 1: Thanks, Eric. Yeah, we have a couple questions, some rapid fire. So first question, are users able to categorize videos, i.e., high-use videos, required versus optional, and have that categorization tied to the overall review?

[00:24:45] Speaker 2: Yeah, so right now, the settings of this service is at the project level of 3Play, which if you're familiar with 3Play, that really just, if you're not familiar with 3Play, that means really like account-level settings. We are exploring and looking to add to this capability to include folder-level organization, so we can start choosing which types of videos go to which types of settings. So there is something that we are building right now and designing to support some of that categorization that is like more BYO categorization versus like 3Play categorization.

[00:25:28] Speaker 1: Thanks. Next question, how does Pulse do with heavily-accented non-native English speakers or lecturers in languages other than English?

[00:25:38] Speaker 2: So right now, Pulse is primarily an English-based service. We specifically, our audio description models are still English-first for right now. Until we get solid on kind of the user experience, adding additional languages will happen. But the Pulse scores are going to illuminate two things. One thing on the caption side, if there are challenges with the source speaker, meaning bad audio quality or accents that are not adequately covered by the AI speaking engines, we're going to pick up the common problems that we run into with AI-generated captions, and we'll surface those scores adequately. So you'll see those probably skewed towards a lower end, which will expect you to remediate those with human review.

[00:26:40] Speaker 1: Great, thanks. And then one last really quick one. Someone asked, can the audit results be downloaded in an XLS file for the whole batch of videos?

[00:26:51] Speaker 2: We are going to have, we are literally like the top of my list in the next month or two to introduce downloading. It will probably be a CSV format for simplicity, but we will introduce the ability to kind of download the data from the dashboard so it's not trapped in 3Play.

[00:27:12] Speaker 1: Great, thanks, Eric. That is all the time we have for today. I know there were some questions we may not have gotten to, but we can answer any additional questions that you have and also show you how you can get started with Pulse. Just visit that link that we shared in the chat and we will help you out. Thanks again, Eric, for giving a really good overview of audio description with Pulse. Thank you everyone in the audience for joining us and asking great questions and just for being here today. Thank you again, everybody, and I hope you all have a wonderful rest of the day.