E-learning Voiceover with Text-to-Speech: A Practical Guide

Text-to-speech is now good enough to narrate most e-learning: compliance modules, software walkthroughs, onboarding and product training. Its biggest advantage isn't the first recording but every update after it, because you can change one sentence without booking a narrator or re-recording a whole module. A human narrator is still worth paying for when a course depends on personality, storytelling or a recognizable presenter.

Below is a practical workflow: deciding when to use AI narration, writing scripts that work aloud, getting timing and pronunciation right, keeping courses up to date, and translating them.

When should you use text-to-speech instead of a voice actor?

Use this as a quick filter.

Situation	Text-to-speech	Human narrator
Content changes often (software, policy, pricing)	Strong fit	Expensive to keep current
Many modules that must sound consistent	Strong fit	Possible, but sessions drift over months
Several languages	Strong fit	One narrator per language
Short deadline	Strong fit	Depends on availability
Flagship course sold on the presenter's personality	Weaker	Strong fit
Emotional storytelling, drama, role-play with nuance	Possible with careful direction	Usually better

A common middle path: a human records the welcome video and the course's emotional moments, and text-to-speech handles the procedural modules that change every quarter.

How do you write an e-learning script that sounds good read aloud?

Most "robotic" narration is caused by the script, not the voice. Text written to be read on a slide is dense; text written to be heard is not.

Rules for writing for the ear

• One idea per sentence. Aim for sentences of 15–20 words. Split anything with more than one comma-separated clause.
• Put the action first. "Click Save to keep your changes" beats "Your changes, which will otherwise be lost, can be kept by clicking Save."
• Say numbers the way people speak them. "About two thousand" is easier to follow than "1,987" unless the exact figure matters.
• Signpost. "There are three steps. First…" helps listeners who can't glance back up the page.
• Don't read the slide. If the narration repeats every word of on-screen text, learners split attention between reading and listening. Multimedia learning research calls this the redundancy effect. Put keywords on screen and the explanation in the audio.

Before and after

Slide-style text:

"Pursuant to policy 4.2, all expense submissions exceeding the threshold must include itemized receipts, failing which reimbursement may be delayed."

Written for the ear:

"If an expense is over fifty dollars, attach an itemized receipt. Without one, we may not be able to pay you back on time."

If you don't have a script yet, Leelo's AI script writer can draft one from your outline, and you can edit it from there.

How long will the narration be?

Professional narration usually runs at about 150 words a minute, and training content often goes a little slower so learners can follow on screen. Use that number to plan.

• A 5-minute micro-lesson is about 750 words.
• A 20-minute module is about 3,000 words.
• A course of 10 modules averaging 8 minutes is about 12,000 words.

Paste a script into the free script length calculator to check its running time before you produce it. If a module runs long, cut words, not speed: speeding the voice up to fit a slot makes it harder to understand.

How do you produce the narration step by step?

Here is the workflow in Leelo's text-to-speech studio, which works line by line.

1. Import the script. Paste it or import a document, then work through it line by line. Each line can be edited and regenerated on its own.
2. Cast the voice. Pick one narrator from the voice library and use it for every module in the course. Consistency matters more than the "perfect" voice.
3. Fix pronunciation once. Add product names, acronyms and jargon to the pronunciation dictionary so "SaaS", "SQL" or your company name are said the same way in every module.
4. Direct the delivery. Slow down a line with a key definition, add a pause before a question, lift the energy for the intro.
5. Keep several takes of tricky lines. Generate a few versions and keep the best; the others stay available if you change your mind. See takes.
6. Add a second voice where it helps. Scenario-based training ("a customer calls to complain…") works better as a dialogue with two speakers in the same project.
7. Export. Download MP3 or WAV per module or per slide and import it into your authoring tool.

Pronunciation entries most courses need

• Acronyms read as letters vs words: "FAQ" (F-A-Q or "fak"), "GDPR" (letters), "NASA" (a word).
• Product names: a brand spelled "Qlik" or "Xero" may be read wrongly without a rule.
• Heteronyms: "lead" the metal vs "lead" a team, "record" the noun vs "record" the verb. Rewrite the sentence or set a pronunciation for that line.
• Units and symbols: "5 ms", "24/7", "°C". Decide how each should be spoken.

How do you keep audio levels consistent across a course?

Learners notice when one module is quieter than the next and they have to keep adjusting the volume. Pick a target loudness for the course and stick to it. Many teams aim for around -16 LUFS for audio played online. Check a few exported files with a loudness meter, such as Leelo's free loudness checker, before you publish.

If you add a music bed, keep it well under the voice and drop it out entirely during dense explanations. Leelo's music and sound-effect lanes sit under the narration, so you can set the mix per section.

How do you update a course without re-recording it?

This is where text-to-speech saves the most time. Say the interface changes and a button called "Submit" is now called "Send for approval":

1. Open the project and find the lines that mention "Submit".
2. Edit those lines (say there are three).
3. Regenerate only those lines. The rest of the module is untouched, so the voice, pacing and levels still match.
4. Export and replace the audio in your authoring tool.

With a human narrator, the same change means booking a session, matching the original mic setup and hoping the voice sounds the same months later.

How do you translate e-learning into other languages?

You have two routes.

Re-voice the script

Translate the script, then produce it with a native voice in each language, for example Spanish, French or German. This gives you the most control and works well for slide-based courses. Leave room in the timing: translated text is often longer than English, so a slide timed to the English audio may need extra seconds.

Dub the video

If the module is a screen recording or a video with a presenter, dubbing translates and re-voices the whole video. You can try it on a short clip with the free video translator first.

What does AI narration for a course cost?

Leelo charges in credits: 1 credit is 1 word in a standard voice, and premium voices cost 1.5 credits a word. Failed jobs give the credits back. A worked example for the 10-module course above (about 12,000 words):

• Standard voice: about 12,000 credits.
• Premium voice: about 18,000 credits.
• Add regenerated lines and alternate takes, and budget perhaps 20–30% more.

That fits comfortably in the Starter plan ($19 a month, 80,000 credits). Dubbing a finished video is priced separately at 5,000 credits a minute. Monthly credits don't roll over, so plan big productions within a billing month or buy a credit pack, which lasts 12 months. Full details are on the pricing page, along with how credits are counted. Commercial use of the audio, including in courses you sell, is allowed on paid plans.

How does narration make a course more accessible?

Audio helps learners who find dense text hard going, including many people with dyslexia (we cover that in text-to-speech for dyslexia). But audio alone can exclude deaf and hard-of-hearing learners. Accessibility guidelines such as WCAG expect captions for video and a transcript for audio-only content. Since your script is the transcript, publish it alongside the audio, and use Leelo's transcription tool to produce text for any recordings that don't have a script.

More ideas for course teams are on the e-learning use case page.

Frequently asked questions

Will learners mind that the narrator is an AI voice?

Most learners care far more about clarity, pace and audio quality than about who recorded it. Problems usually come from scripts written for the page, mispronounced terms and inconsistent levels, all of which you can fix. Be open about it if asked; there's no need to hide that the narration is synthetic.

What's the best speaking speed for e-learning narration?

Around 150 words a minute is a good starting point, slightly slower for technical definitions or learners working in a second language. Adjust the speed per line rather than across the whole module, so key points get more time.

Can I use AI narration in a course I sell?

On Leelo's paid plans, commercial use of the audio is allowed, which covers paid courses and client work. The Free plan's 1,000 one-time credits are for trying the studio. Check the terms of the platform you sell on as well.

Which audio format should I export for my authoring tool?

Most e-learning authoring tools accept both MP3 and WAV. Use WAV if you plan to edit the audio further, and MP3 for smaller files when you import directly. Export one file per slide or per section to make later updates easier.

How do I keep the same voice across courses made months apart?

Use the same voice, the same delivery settings and the same pronunciation dictionary for every module. Because the voice is synthetic, it doesn't age or change between sessions the way a human recording setup can. If you want a voice no one else uses, design one or clone a consenting presenter's voice.