Back to Blog
Education

Text to Speech for E-Learning: A Course Creator's Guide

Text to speech for e-learning: when synthetic narration works, how to keep courses accessible, what authoring tools offer and what narration costs, as of 2026.

10 min read
By OndaLabs Team

Yes, text to speech for e-learning works for most narrated modules. It lets you regenerate narration after a script change and keep one voice across a long course. Evidence on synthetic voices is limited but encouraging. Captions, transcripts and a check of your licence and local rules still apply.

What the research says about AI voices and learning

The most relevant study is Dincer (2022), published in the Journal of Pedagogical Research. Fifty-one university students watched the same video with one of three voices: a traditional text to speech voice, a modern text to speech voice, or a human narrator. Post-test means were 42.67 for the traditional voice, 54.12 for the modern voice and 60.56 for the human voice.

Both the modern voice and the human voice scored significantly higher than the traditional voice. The modern and human groups did not differ significantly from each other (p = .200), and mental effort did not differ significantly between them either. The modern voice came from Microsoft Azure and the traditional voice from Google Cloud, all in American English.

Treat this as one small study. The same paper also summarises older work on Mayer's voice principle, which holds that people learn better from a human voice than a machine voice. It cites Craig and Schroeder (2017), who found no significant difference between a modern machine voice and a human voice. That is a secondary summary, so read the original studies before you rely on it.

Slides matter as much as voices. Yue, Bjork and Bjork (2013) tested narrated PowerPoint lessons with college students. On-screen text identical to the narration impaired learning, and a small discrepancy between text and narration helped. Learners preferred identical text anyway. That is why you should not put a permanent, word-for-word copy of the narration on every slide. Captions are a different thing: a separate track that viewers switch on, and accessibility rules still expect them.

Which authoring tools include text to speech?

Authoring tools are adding narration features, but the details vary and change quickly. The table lists what each source says as of October 2026.

| Tool | What it offers | Source | |---|---|---| | Adobe Captivate | Captivate 12.5 added AI voices for text to speech. In the Audio Inspector you generate speech, pick a language and voice, adjust pacing and pronunciation, then generate audio. Different voices can be assigned to different captions. | Adobe eLearning blog, January and May 2025 | | Adobe Captivate voices from ElevenLabs | ElevenLabs says its voices are available "directly in the Adobe Captivate app" and describes more than 150 AI voices across 10 languages. This is a vendor statement from June 2025. | ElevenLabs blog | | Articulate Rise 360 | Can generate text to speech narration since the July 2025 update, according to Articulate's help pages as summarised in search. Learners press play and can change speed, but cannot download the audio. Custom voice upload is not supported. | Articulate community help pages, search summary | | Articulate Storyline 360 | Has a voice library that can be filtered by gender, age and accent, per Articulate's help pages as summarised in search. | Articulate community help pages, search summary |

Articulate's 360 pricing page does not mention text to speech, so do not assume that a plan includes it.

How do you keep narrated courses accessible?

Narration is audio, so it falls under the same accessibility criteria as other audio content. WCAG 2.2 became a W3C Recommendation on 5 October 2023 and was updated on 12 December 2024. It keeps the criteria that matter most for narration from WCAG 2.1.

| Criterion | Level | What it means for narrated courses | |---|---|---| | 1.2.1 Audio-only (Prerecorded) | A | Provide an alternative, such as a transcript, that gives equivalent information and identifies speakers. | | 1.2.2 Captions (Prerecorded) | A | Prerecorded audio in synchronized media needs captions. Narration over slides or video generally needs them. | | 1.2.3 Audio Description or Media Alternative (Prerecorded) | A | Needed only when the visual information is not already conveyed in the audio. The guidance says audio can be synthetic but does not rule on synthetic voices. | | 1.4.2 Audio Control | A | Audio that plays automatically for more than 3 seconds needs a way to pause, stop or change its volume. | | 3.1.1 Language of Page | A | A declared page language lets screen readers load the right pronunciation rules. |

Two practical points follow. First, do not autoplay narration. Start it when a learner presses play, because sound that starts on its own can interfere with a screen reader. Second, a clearly labelled media alternative for text can be an exception to the captions requirement, but only if it presents no more information than the page text. The WCAG 2.2 captions guidance does not address narrated slides directly, so ask your accessibility lead how to apply it to your slides.

In the United States, the ADA Title II web rule covers state and local governments, including public schools, community colleges and public universities. It requires WCAG 2.1 Level AA. An interim final rule published in the Federal Register on 20 April 2026 set compliance dates of 26 April 2027 for governments serving 50,000 or more people, and 26 April 2028 for smaller governments and special districts. Older articles still show earlier dates. The ADA guidance does not discuss course content specifically, so check how your institution applies it. Federal agencies follow Section 508, which requires WCAG 2.0 Level A and AA for electronic content.

This is general information, not legal advice.

What does e-learning narration cost?

Vendors price speech by the character. The table compares published list prices per one million characters, as of October 2026.

| Service | Published price per 1M characters | Source | |---|---|---| | Amazon Polly Standard | $4.00 | AWS pricing | | Amazon Polly Neural | $16.00 | AWS pricing | | Amazon Polly Generative | $30.00 (speech requests only) | AWS pricing | | Amazon Polly Long-Form | $100.00 | AWS pricing | | OpenAI tts-1 | $15.00 | OpenAI pricing | | OpenAI tts-1-hd | $30.00 | OpenAI pricing | | Azure AI Speech Neural (East US) | $15.00 | Azure retail prices | | Azure AI Speech Neural HD (East US) | $22.00 | Azure retail prices | | OndaLabs Onda Lite on Starter | About $52.94 (derived from the $9 plan; credits are shared with other tools) | OndaLabs plans |

Amazon's Polly page says that 1 million characters is about 23 hours and 8 minutes of speech. If you use that conversion as an assumption, a one-hour script is about 43,200 characters. At Polly Neural rates, that is about $0.69 of speech. This is a rough estimate built on one vendor's conversion, so measure your own script before you budget.

OndaLabs plans, in monthly USD:

  • Free: $0, 2,000 credits (about 2 minutes), Onda Lite
  • Starter: $9, 170,000 credits (about 189 minutes), Onda Flash, commercial use
  • Creator: $22, 400,000 credits, 3 custom voices, music generation
  • Pro: $49, 900,000 credits, 10 custom voices

Onda Lite uses 1 credit per character, so a 43,200-character hour would use about a quarter of the 170,000 credits on Starter, if the same conversion held for your script.

A workflow for narrating one module

Use this sequence for a single module. It keeps re-recording small and makes corrections cheap.

  1. Write for the ear. Use short sentences, and spell out numbers, dates and acronyms the first time they appear.
  2. Split the script by slide. Regenerating one slide is much quicker than rebuilding the whole course.
  3. Choose the language and voice. Use one voice for the whole course unless you have a reason to change it.
  4. Generate the audio and listen with the slide on screen, not in isolation.
  5. Check names, technical terms and pronunciations. Respell them in the script and regenerate only that slide.
  6. Add captions or a transcript. Speech to text can draft one from narration you already have, and a person must correct it before publishing.
  7. Set narration to start on a learner action, not when the page loads.

Where OndaLabs fits

OndaLabs is an AI voice studio. It creates audio files. It does not sit inside your authoring tool, so you import the files into Captivate, Storyline, Rise or your learning platform yourself. Do not expect a direct integration.

  • Text to speech: 30 prebuilt voices, usable in every supported language, with dedicated pages for 37 languages. Start with text to speech.
  • Two-speaker dialogue: useful for scenario-based modules, such as a manager and an employee role-play, in the dialogue studio.
  • Custom voices: available on Creator and above, cloned only from a recording you have permission to use, so a course can keep the instructor's voice.
  • Speech to text: 50 credits per minute, to draft a transcript from narration you already recorded. A person still reviews it, using the speech to text tool.
  • Commercial use is included from the Starter plan at $9 a month.

If you need the wider accessibility picture for a site, read our text to speech accessibility guide. For localising a course into another language, see language learning with text to speech.

Frequently asked questions

Can I use text to speech for e-learning courses?

Yes, for most narrated modules, provided you check the licence for the voice and tool you use. Synthetic narration is practical when scripts change often, because you can regenerate a single slide. The learning evidence is limited, so pilot one module and collect learner feedback before you narrate a full course.

Do learners learn as well from AI voices as from human narrators?

The best-known study, Dincer (2022) with 51 students, found no significant difference between a modern synthetic voice and a human voice on post-tests. It is one small study on one topic.

Do I still need captions and transcripts if I add narration?

Generally yes. WCAG 2.2 expects captions for prerecorded audio in synchronized media, and a transcript is the alternative for audio-only content. Captions are a viewer-controlled track, which also avoids the redundancy problem that word-for-word on-screen text causes. Confirm the exact requirement with your accessibility lead.

Should on-screen text match the narration word for word?

No. Yue, Bjork and Bjork (2013) found that on-screen text identical to the narration impaired learning. Keep slide text short and complementary rather than a copy of the script, and use captions for accessibility.

How much does AI narration cost for a one-hour course?

It depends on the length of your script and the engine you choose, so there is no fixed figure. Using Amazon Polly's own conversion of about 43,200 characters per hour, Neural speech would cost roughly $0.69 at $16 per million characters. Measure your own script before you set a budget.

Can I use AI voices in a commercial training course?

Check each vendor's licence terms before you sell a course. On OndaLabs, commercial use is included from the Starter plan at $9 a month. Ask your legal adviser if the course will be sold.