Voxsmith

I Used to Voice Some of Our eLearning. Here’s Why I Use AI Now (And When I Don’t)

by

I’ve been doing professional eLearning voiceover work for years. I was the voice of an entire product line’s customer-facing training content. Hundreds of slides, dozens of courses, all recorded in my home studio using Hindenburg Pro.

So when a VO professional challenged me this weekend about using ElevenLabs AI voices, I got it. I’m on their side. But I also have a giant “but” – and it’s rooted in the economics and workflows of corporate eLearning that nobody wants to talk about.

What Changed (And It Wasn’t the Technology)

I still do voiceover work. In fact, I do better voiceover work now. The difference? My time is now reserved for high-value content: HR training, audit-related courses, compliance content – things with longer shelf lives and larger organizational impact.

The routine product training that changes quarterly? The quick-turnaround customer education modules? The content that gets revised six times before launch? That’s where AI voices entered the picture.

Not because AI replaced me. Because someone finally did the math on my time.

The Economics Nobody Talks About

Here’s the reality of most corporate eLearning budgets:

What I’d love to have: $50,000/year for professional voiceover work across all content.

What I actually get: $1,500.

When you’re working with that kind of budget, and your content changes twice a year minimum (plus add-ins, plus revisions, plus the inevitable “can we just tweak this one slide?” requests), the traditional VO workflow becomes impossible.

It’s not that human voiceover isn’t worth it. It’s that the content update cycle and the budget reality don’t align. You end up with a choice: outdated narration or no narration at all.

What AI Voices Actually Enable

With AI voices handling the high-churn content, we can actually:

  • Create more engaging content within constraints – Narrated courses consistently outperform silent slides for adult learners, but most budgets can’t support full VO. AI makes engaging, narrated content financially viable.
  • Keep content current instead of letting narration go stale because updates are too expensive
  • Deliver faster turnarounds for customers who need training materials quickly
  • Iterate more freely during development without revision fees for one slide change

This isn’t about replacing voice talent. It’s about making narrated eLearning financially viable for content that needs constant maintenance.

When Human VO Always Wins

AI voices are a tool, not a replacement for actual voice acting. Here’s when I still reach for the microphone:

  • High-stakes content – HR policies, audit training, anything with serious compliance implications
  • Long shelf life – Content that won’t need updates for 2+ years deserves the investment in human performance
  • Brand-critical material – First impressions, flagship courses, anything customer-facing that represents the company
  • Emotional nuance – When you need empathy, warmth, or genuine human connection

For these projects, the extra time and budget are absolutely worth it. The personality, inflection, and authenticity of a human voice matter.

The Tool I Built for the Middle Ground

Working in both worlds – professional VO and AI voice implementation – I kept hitting the same workflow problem: getting AI-narrated PowerPoint decks produced was still too manual and time-consuming.

That’s why I built Voxsmith. It automates the tedious process of converting PowerPoint speaker notes into narrated presentations using ElevenLabs voices. What used to take hours of exporting, converting, and syncing now happens in minutes.

It’s designed specifically for L&D professionals who need to produce narrated content quickly, iterate frequently, and stay within realistic budgets.

The Bottom Line

I’m not anti-voice talent. I am voice talent. But I’m also a pragmatist working in corporate learning environments where budgets are real and content changes constantly.

AI voices let us say “yes” to narration on projects that would otherwise go silent. They free up budget and time for human VO where it actually matters. And they let me focus my own voiceover work on content where my experience and craft make a real difference.

That’s not disruption. That’s just using the right tool for the job.

Don Burnside is a senior learning specialist at Axway and creator of Voxsmith, a PowerPoint narration automation tool for eLearning developers.

Voxsmith Demo Update

by

The UI is done. All of the features are installed and working. The only think left to do is test. A lot of testing! I have 3 decks with almost 200 slides to process this week, and another 3-5 decks next week. Once done, I’ll make sure all of the remaining bugs get squashed and get it shared!

If you are interested in testing Voxsmith, please comment below!

=====

Not related, but you know when you finish something and you are a little surprised that it’s something you created? This video. This app. All part of that. Holy crap!

Voxsmith Update:

by

Leading into the home stretch before release; it’s bug squashin’ time! I’ve put the current dev version of Voxsmith to work this week and expect to really be able to put it through its paces! I’ve already found 2 bugs and was able to correct 1 and redirect the other.

Voxsmith, at its core, attaches audio to PowerPoint slides, programically, automatically. This is amazing when you run Voxsmith as the first part of your design process, when it’s just the raw slides from your curriculum team or SME. Easy peasy. This is what the original Narration Generator did. Just make sure not to use it on slides with animations; it would do bad things.

Researched this and found a bug in MS PowerPoint COM implementation where it literally re-writes animations when an audio file is attached via COM. This is problematic, since I also want to be able to use Voxsmith during my design phase, when I need to correct a typo or adjust some timing. And I don’t want to drag on drop.

To work around this limitation and to make sure I do not get the angry “YOU BROKE ANIMATIONS ON MY 200 SLIDE DECK I HAVE BEEN WORKING ON FOR 6 MONTHS!” emails, I developed an animation detect, backup, restore, redirect:

When Voxsmith runs, it first digs into your deck to grab the script and check for animations, if there are any found that aren’t text-boxes, it backs them up, removes them, attaches the audio and restores the animations. It’s amazingly elegant to watch!

This doesn’t work with text boxes. PowerPoint gives special rules to text boxes, making them mostly untouchable in the way I want to back them up. No, I tried that. Tried that too. Not to worry. Instead, when Voxsmith finds these animation types, it will still process your narration, but it will only go to your Output folder and it will not attach to the slide. Drag and drop rules apply, because animations.

I’ve also started building short-codes for Voxsmith that will let you do certain things. To start, it’s a simple way to get Voxsmith to “Read the slide” that is slightly faster than copying and pasting the slide text into the notes. I will be using short codes to facilitate deck splitting (“chunking”), Making Voxsmith the complete tool for dealing with learning video creation from PowerPoint.

Wait until I start teasing you about Voxsmith 3. It is going to be wild!

Where does Voiceover belong in the L&D Workflow?

by

For years, the instructional design workflow has been a predictable relay race: build in PowerPoint, export to Storyline, then layer in narration once everything’s locked down. The process works — it’s just painfully slow.

Voiceover always becomes the bottleneck. Either you record your own scratch tracks (and pray you never have to re-record them), or you wrangle text-to-speech tools in a browser tab, copy the clips back into PowerPoint, test timing, export again, and then import to Storyline. That’s a nine-step loop that kills creative momentum.

But here’s the thing: PowerPoint already sits at the center of your design process. It’s where pacing, tone, and flow come together long before your first Storyline trigger ever fires. So why not let your slides speak while you’re still designing?

That’s the shift Voxsmith makes possible. Instead of treating narration as a post-production step, you can generate professional-sounding voiceover directly from your PowerPoint speaker notes — instantly. That means you can build, test, and hear your content during early drafts. You can adjust phrasing, check pacing, and make real design decisions based on sound, not just text.

Once your deck feels right, you import it straight into Storyline — fully narrated, timing baked in, ready for sync and polish. It’s faster, cleaner, and way closer to how real courses get built in the wild.

Audio used to be the last thing you added. Now it’s the first thing that helps you think. The demo is live and I am looking for testers!

Introducing Voxsmith

by

I am pleased to show you my latest app, 100% functional and just a little rough around the edges. This is how to you attach Elevenlabs.io audio to your PowerPoint slides. Welcome to Voxsmith!

I’m looking for a few testers before this becomes available for everyone! Comments are open if you or someone you know could use this tool, let me know!

Using the right tool for the job

by

Updated UI for Voxsmith

Boy howdy am I learning a lot about robots, about what kind of capabilities the robots have, and how absolutely maddening they can be to use, if you are using the wrong one. Which has really been my experience using both Copilot and ChatGPT for more “advanced” task, like creating an advanced application that generates audio narration from Power Point speakers notes using Elevenlabs, for example.

You can go prompt after prompt and get into a real rhythm and knock out some serious work! Until, who knows why, they suddenly get into what I call a ‘dumb loop’, where you request something, and it says it’s going to do the thing as soon as you ask, so you ask, and it tell you it will do the thing after you ask, and so on. This happened to me frequently in both Chat and Copilot. I tightened up my asks quite a bit, and I would still get this.

Then, for coding, everything is going great, until a row is 3 spaces off, or there is an extra comma. Those are the things that take hours to resolve. I had an issue like this yesterday with Voxsmith. It was a regression that potentially messes up animations in Powerpoint. Spent over 7 hours yesterday with Chat and Copilot, and neither of them could correct it, even after I showed both of them the code where it worked in a previous version.

So, I switched to Claude. On the free version, I uploaded my app files, it did code review, I said bug report and in 3 revisions it got it right. Actually 2, but the third I was being pedantic. Call it 20 minutes. My mind was blown!

I signed up for a paid account immediately (and cancelled my ChatGPT paid account) and it did not mess around, getting to work right away with code clean up a complete UI refresh and a bunch of bug fixes. This was done in less than 8 hours total!

This post was going to be my first Voxsmith demo. If I’m honest, I’m really glad everything fell to shit on Monday with Chat and Copilot; then I wouldn’t have found Claude and wouldn’t have made the progress that I have made, because now I think I am using the correct tool for the job

App Updates

by

The Voxsmith v2.1x track is almost complete. Only some final security touches and a feature implementation and it will be ready to rock. Demo video I expect should be ready sometime in the next couple of weeks.

And today I stuck a flag in the ground for my next app: FlowPath. There is a spec. There is a Powerpoint of mock screen layouts. There is a v0.0.1 that actually does the thing, in a very limited fashion. And I think it’s going to be pretty neat!

FlowPath is a How-To creator. Desktop only, cross platform, local first. It allows the user to create how-to’s quickly with screen shots and steps, saved to their local computer with the ability to sync to a team folder. Those How-To’s can then be viewed in the app or exported, shared or printed. It’s gloriously simple. And it will tie in with Voxsmith at some point, being able to create audio how-to’s and video walk-thru (future version).

I was going to break up with developing earlier this week. I was having some issues with my meds, which put me in a very bad place. Some naps and med changes later, I re-examed the project with new, slightly wearier eyes. And I started working.

But working in my way. I’m not a developer. At best, I could be called a “web designer”, but ew, no. What this really means is I am very visual. I have to see a thing before I can make it do what it needs to do. Backwards. Still, I started researching with ChatGPT, having it generate some wireframes. Those don’t quite get close enough for me, so I opened a blank powerpoint deck and quickly laid out the 4 primary screens of the app.

Did I mention gloriously simple?

A Home Screen, a creation screen, a setup screen and a reader, which is reused from the setup screen.

It will do its thing, it will do it quickly and without fault. Hopefully. And, it will do it on Windows, Mac and Linux. No, really. Comments are open if you have questions about Voxsmith or FlowPath.

 

Voxsmith 2 – The Beginning!

by

After months of building, debugging, and arguing with LLMs, Voxsmith is officially real. It’s at the point where I could stop right now and be proud of what it does.

So what does it do? It connects your PowerPoint deck to ElevenLabs and automatically creates audio narration for your slides. On the first pass, it even attaches the audio directly to each slide. You can process the entire deck, a range, or just one slide–perfect for quick edits after the initial run.

V2.12 got the option to close the deck automatically when the process finishes–so you can kick off a session and let Tomorrow You handle it.

A few more updates are coming to make version 2 truly complete. The first is Draft Mode, which lets you test your slides with a local OS voice (or a better one if available) so you can tweak and time things without burning ElevenLabs credits.

We’re also locking down API security, moving away from plain-text keys and paving the way for multi-key support in future releases.

I’ve tested this thing, and it’s fast. A full 27-slide deck took three minutes flat. Where are my PowerPoint-to-ElevenLabs copy-and-paste wizards? This will save you days–unless you’re already living that macro life.

All of these remaining updates will bring Voxsmith to v2.15. For now, it’s Windows only, with full Mac support coming in version 3.

If this sounds useful in your workflow, drop a comment, I am looking for beta testers!