Facts, menus and product details checked in November 2026.
To narrate slides with an AI voice, write what you would say over each slide, one short paragraph per slide, and turn each paragraph into its own audio file with a text-to-speech tool. Then put each file on its slide (in PowerPoint: Insert > Audio, then Start: Automatically) and, if you need a video, give each slide as many seconds as its audio before you export. One file per slide is the whole trick: when a fact changes, you re-voice one paragraph and swap one file, and nothing else moves.
But the tools do less than their pages suggest. PowerPoint has no button that turns your notes into a saved voice, and its video export gives a slide its timing, not the length of the audio on it, so a voice can be cut off mid-sentence. Google can turn a deck into a narrated video, but only on eligible paid Workspace editions or Google AI plans, not on the free Education Fundamentals edition. And in Kukarella, where we make the audio, you bring the text in as a PDF or as plain text: it does not open a PowerPoint file, and it has no PowerPoint add-in. Each section below says what holds for any tool, then what to do in Kukarella.
In this guide:
- Is an AI voice good enough to teach with?
- Step 1: write the narration for the ear.
- Step 2: make the audio. Which tool?
- How to make the audio in Kukarella.
- Step 3: put one file on each slide.
- Step 4: export a video without cutting the voice.
- Captions: who else is listening.
- When a fact changes: re-voice one slide.
- Can Kukarella turn the narration into a video?
- Frequently asked questions.
Is an AI voice good enough to teach with?
Short answer: in most recent studies, yes: people learn about as much from a modern AI voice as from a human recording. But they still slightly prefer a human, and an odd-sounding voice can cost you their attention.
Narration itself is the strongest idea here. Richard Mayer, whose experiments underlie much of the advice on multimedia teaching, counts the "modality" effect, explaining a picture with spoken words rather than printed ones, as winning 18 of 19 tests, with a median effect size of 1.00. That is a large effect for education research. In plain words: a slide with a picture and a voice teaches better than the same slide with a paragraph of text under the picture.
The voice question has changed since the early studies. In 2003, Mayer, Sobko and Mautone found that learners did better with a human voice than with a synthesized one, and Mayer cites it for his "voice principle", which held in 6 of 7 tests. The machine voices of 2003 were the robotic ones you remember. With modern voices the picture is different:
- Craig and Schroeder (2019) compared an old synthetic voice, a modern one and a recorded human narrating an instructional animation. The modern voice was "in most respects" no different from the human on learning, perceptions or cognitive efficiency.
- Netland and colleagues (2025), a study from ETH Zurich with 447 participants and four short course videos: AI-made videos taught as well as human-made ones on the exam, while the human videos had a small edge in how much people enjoyed them. Torbjørn Netland wrote that the human videos took several days to make and the AI ones a few hours.
- The counterweight, Jing and colleagues (2024): in a small eye-tracking study of 30 students, a human recording beat two AI voices, a formal one and a "cute" one, on learning and attention.
No study I found pools the modern-voice results, and each one tests a single lesson. So the fair conclusion is: pick a clear, natural, adult voice, listen to it before your learners do, and let the content carry the lesson. Choosing that voice is its own question: How to Choose the Right AI Voice.
There is also a reason to narrate at all. In 2003 the board that investigated the loss of the space shuttle Columbia wrote about NASA's "endemic use of PowerPoint briefing slides instead of technical papers". A slide holds headlines; the explanation was somewhere else. Narration is where the explanation goes.
Step 1: write the narration for the ear
Short answer: write what you would say, not what the slide says, about 150 words a minute, one paragraph per slide. But do not delete every word from the slide: short phrases on screen are fine.
The most common mistake is to voice the bullets. Listeners hear a list of fragments, and they read the same fragments faster than the voice says them. Mayer's own worked example is a short narrated slideshow on how a solar cell works: short printed phrases on the slides, and a conversational voice that says, for example, "Let's see how a solar cell works." His "redundancy" count is the reason not to go further than that: leaving out on-screen captions that repeat the narration wins only 8 of 12 tests, with a small median effect of 0.10. So the rule is "don't read a wall of text aloud", not "no text on slides". Mayer notes that printed words can help when the lesson is in the learner's second language.
Here is one slide of a staff-training deck, before and after.
On the slide: Report a near miss within 24 h · Use the near-miss form · Tell your supervisor the same day.
Narration that reads the slide (do not do this): "Report a near miss within 24 h. Use the near-miss form. Tell your supervisor the same day."
Narration written for the ear: "Nobody was hurt. So does it still need a report? Yes. A near miss is the cheapest warning you will ever get, and we want to hear about it within a day. Fill in the near-miss form, it takes about five minutes, and tell your supervisor before you go home. They are not looking for someone to blame. They are looking for the next accident."
Here is that narration in a British English voice from the Voice Library: 60 words, 23 seconds.
The near-miss slide, written for the ear (English, UK)
The second version says why, answers the question a learner is already asking, and spells "24 h" as "within a day", because a voice reads what is written. A few more rules, each from the same habit of writing for a listener:
- One idea per sentence, and sentences short enough to say in one breath.
- Say what the chart means, not that there is a chart: "Sales doubled in March", not "As you can see on this slide".
- Write numbers, dates and abbreviations the way they should sound, and listen to every name.
- End each paragraph where the slide changes. If a paragraph runs past about 100 words, split the slide.
- Keep each video short. In 6.9 million viewing sessions on edX, Guo, Kim and Rubin found that the median time people watched was at most six minutes, however long the video was. That measures watching, not learning, but it is a good reason to cut a long deck into several short ones.
How long will it run? Course videos in that study were spoken at 156 words a minute on average, and the audiobook marketplace ACX estimates about 9,300 words for a finished hour, about 155 a minute. So a 60-word paragraph takes about 23 seconds. The planner below does the counting for your whole deck.
Paste your notes into it, one slide per paragraph, and it gives you each slide's length, the seconds to set for it in PowerPoint, and the lines a voice is likely to trip on.
If your starting point is a document rather than slides, the same writing rules are in PDF to Script AI: How to Turn Any PDF Into a Voiceover Script.
Step 2: make the audio. Which tool?
Short answer: if your school or company has a paid Google Workspace edition, or Microsoft 365 with Copilot Chat, try the built-in route first. But each one has limits, and a separate voice tool gives you the most control over the words and the voice.
Here are the options as they stand in November 2026, each with its "but".
PowerPoint on its own
What you get: your own voice, recorded slide by slide (Record tab). The Speak command and Immersive Reader read text aloud, but neither saves an audio file.
But: no AI voice you can keep. Recording yourself means a quiet room and a new take for every slip.
Google Slides with Google Vids
What you get: File > Convert to Video turns the deck into a video with an AI voice. You can choose Replace with speaker notes so the voice reads your notes, and each scene lasts as long as its voice.
But: it needs an eligible paid Workspace or Google AI plan. Google's April 2026 list includes Education Plus, not the free Education Fundamentals edition. AI narration works for up to 45 slides and in eight languages.
Microsoft 365 with Copilot Chat
What you get: in the Copilot app, Create, then a video from a PowerPoint file, made and edited in Clipchamp.
But: only with a work account (Microsoft Entra ID) on a Microsoft 365 plan with Copilot Chat, and your administrator must have turned Clipchamp on. Microsoft's page says the videos use stock footage.
A video service that reads speaker notes (Narakeet and others)
What you get: you upload the PowerPoint file; the speaker notes become the script and you get a video back. Narakeet sells 30 minutes for $6.
But: the result is a video, not your deck. Narakeet's own pages say animations, transitions and embedded videos are not supported, and its free tier allows no commercial use. A change means building the video again.
A course authoring tool (Articulate Storyline 360, iSpring)
What you get: AI voices inside the course you are building, one clip per slide, published for a learning platform.
But: the price is for course teams. Articulate 360 AI is listed at $1,449 to $1,749 a seat a year.
A text-to-speech tool such as Kukarella
What you get: one audio file per slide, from text you control, in a voice you chose from a large library, ready for PowerPoint, Google Slides, Keynote or an authoring tool.
But: you place the files on the slides yourself. Kukarella does not open PowerPoint files, and Text to Voice makes no caption file: for an SRT, upload the finished audio to TranscribeHub, our transcription app.
Some of our best customers work exactly this way: they make the narration in Kukarella and assemble the course in Articulate. For the course around the slides, one voice for every module and updates without re-recording, see How to Narrate an E-Learning Course With AI.
How to make the audio in Kukarella
Short answer: bring the text in, make it one paragraph per slide, choose one voice, press Play, then Download > By paragraph. You get a ZIP with one numbered audio file per slide. But the numbering starts at 0, not 1: the first file is 0 (00 in a deck of 10 to 99 slides), and it goes on slide 1.
- Bring the text in. Open Text to Voice and paste your speaker notes in, or export the deck as a PDF and upload it with the Media button. It takes PDF, Word (DOCX), plain text, and JPG or PNG pictures, up to three files at once. A slide that is only a picture still gives its text: pages with no text layer are read by an AI image reader, up to 20 of them in one PDF. But a page with any typed text, even a slide number, is read from that text only, not from the words in its pictures. And a PDF arrives one paragraph per line: a title, each bullet and each wrapped line become separate paragraphs. Join them until each slide is one paragraph: the planner above does it for you if you paste the PDF's text with an empty line between slides, then copy the result. A project holds up to 500 paragraphs.
- Or let the AI assistant write it. Attach the PDF to KukaGenie, the AI assistant (up to 10 files, 20 MB each), and choose the ready request Your slides, spoken with no bullet points. It writes a spoken script from the slides. But read every line it writes before you voice it: it writes from what is on the slides, and it can get a fact or a figure wrong.
The assistant's ready request Your slides, spoken with no bullet points, under From your own file.- Choose one voice for the whole deck. There are 2,000+ voices, and the Voice Library filters them by language and accent, gender and age. A Text to Voice project is read by one voice from start to finish, which is what a deck needs. Customers tell me the same thing: once they pick a voice for their course, they don't want to change it. If you want two characters to talk on a slide, write that part in DialoguesAI instead.
- Add the pauses. Our users very often set pauses between paragraphs themselves: their learners need a moment to take in each point. You can put a pause anywhere, from half a second to 90 seconds, or set the pause between paragraphs for the whole project. Speed and pitch work on every account; Say As (to spell out letters or read a number as an ordinal), volume and emphasis come with paid plans. But not every voice takes every effect: the Voice Library's Effects filter shows which do.
- Press Play and listen. The cost is shown before you press it. Most voices cost 1 credit a character; a few cost 2, and on many voices the pauses and other effects add characters to the count. A new free account starts with 30,000 characters, enough for about half an hour of narration in a voice at 1 credit a character. But audio made on the free plan has no commercial rights; for a course you sell or use at work, use a paid plan. Voicing a 20-minute module usually takes one to three minutes of waiting.
- Download one file per slide. Press Download at the top of the project, choose MP3 or WAV, then By paragraph, and press Download at the foot of the box. You get a ZIP with one file per paragraph, named with its number, the first 20 characters of the paragraph (without punctuation) and the voice's name, for example 03. Nobody was hurt So d-[voice name].mp3. But the numbering starts at 0 (00 or 000 in a longer deck), the files sit in a folder inside the ZIP, and a paragraph you did not voice is skipped, so its number is missing. Check that the count of files matches your slides.
The Download box: MP3 or WAV, then One file or By paragraph.Step 3: put one file on each slide
Short answer: insert each slide's file on that slide, set it to start automatically and hide its icon. But do not put one long file on the first slide and let it play across the others.
One long file seems easier, and it breaks at the first change. The audio runs on one clock and the slides on another, so a slide added, removed or moved, or one sentence re-recorded, puts every later slide out of step. A file per slide keeps each slide on its own.
| Tool | How to add a file to a slide | Watch out |
| PowerPoint for Windows | Insert > Audio > Audio on My PC; then Playback > Start: Automatically and tick Hide During Show | Microsoft recommends M4A (AAC); MP3 and WAV also work |
| PowerPoint for Mac | Insert > Audio > Audio from File; then the same Playback settings | Test the exported video, see Step 4 |
| PowerPoint for the web | Insert > Audio; then Audio Options | At most 16 MB a file, so prefer MP3 to WAV |
| Google Slides | Insert > Audio, from your Google Drive only | MP3 or WAV; viewers need access to the Drive file |
| Keynote | Insert Media, or drag the file onto the slide; Format > Audio | No AI voice built in |
A minute of WAV takes about 5 MB or more, a minute of MP3 about 1 MB, so for a long deck download MP3. If you build in an authoring tool such as Storyline, the clips go in the same way, one per slide. One of our customers, a public body in Quebec that makes its training in French, works this way: the team downloads the MP3 files and inserts them into their Storyline projects, which they describe as quite easy.
Step 4: export a video without cutting the voice
Short answer: set each slide's time to the length of its audio plus a second, then File > Export > Create a Video. But test one slide first, especially on a Mac.
PowerPoint's video export does not look at the audio on a slide. It gives each slide its recorded timing or, if there is none, the number of seconds in Seconds to spend on each slide, 5 by default. Microsoft's page states only that default; user forums and the vendors who sell fixes report the audio being cut when the slide is shorter than its file. The fix is per slide: Transitions > After, and type the audio's length plus a second or two. The planner above gives you that number for every slide.
Then File > Export > Create a Video, choose the quality (from 480p to 4K) and Use Recorded Timings and Narrations, so the After times you typed are used instead of the default seconds, and save as MP4. On the Mac, the route is File > Export with MP4 as the format. Microsoft's page for PowerPoint for Mac says that recorded narration plays in the exported video but other embedded media files won't, so an inserted audio file may come out silent. That page may be older than your version: export one slide, play it, and only then export the deck.
Google Slides has no video export of its own; its route to a video is File > Convert to Video, in Google Vids, above. In Keynote, File > Export To > Movie.
Captions: who else is listening
Short answer: a narrated video needs captions, and the narration you wrote is already their text. But do not count on PowerPoint to carry captions into a video, and Text to Voice does not make a caption file. TranscribeHub, our transcription app, makes an SRT from the finished audio, but not a VTT.
A voice helps the people who struggle with print: the World Health Organization counts at least 2.2 billion people with a vision impairment. It shuts out the people who cannot hear it. The current web accessibility standard, WCAG 2.2, asks for captions on all prerecorded audio in a video and a text alternative for audio on its own. CAST's guidelines for Universal Design for Learning put the same idea for teachers: "Share information in more ways than images and text alone."
- Inside the deck: PowerPoint for Windows (version 2303 and later) and for Mac (16.71 and later) can attach a caption file to inserted audio: select the audio, then Playback > Insert Captions. It takes WebVTT (.vtt) and, in newer versions, SRT.
- In the video: Microsoft says Compress Media drops captions, and a video made from the deck leaves out the captions of embedded videos; do not count on captions on audio either. Upload a caption file next to the MP4 on your learning platform or on YouTube instead.
- The cheapest alternative: publish your narration text with the deck. It is the transcript, and for audio-only material it is what WCAG asks for.
To make a caption file from the finished audio or video, see How to Make an SRT or VTT File From a Video.
When a fact changes: re-voice one slide
Short answer: edit the paragraph, voice it again, download that one paragraph and replace the file on its slide. But keep the same Playback settings and timing, or the slide will behave differently.
This is where a voice made from text pays for itself. A recorded voice needs the same room, the same microphone and the same voice on a different day. Here, you change the sentence, press Play on that paragraph and download it on its own as MP3 or WAV: each paragraph has its own download button. The file is named after the project, so rename it before you swap it in. In PowerPoint, delete the old audio on the slide, insert the new file, set Start: Automatically and Hide During Show again, and correct the slide's After time if the length changed. Services that build a whole video from your deck make you build the whole video again.
Can Kukarella turn the narration into a video?
Short answer: yes, but not with your slides. Make video in Kukarella turns a voiced Text to Voice project or dialogue into an MP4 with a picture for each moment and captions, and the pictures are drawn by AI from your text. It cannot use your slide images.
What it adds: a short video when you have no slides at all, or a scene to open a lesson, in 16:9 for a screen or 9:16 for a phone, with captions burnt in. The pictures are drawn by AI, so check each one, and nobody's lips move.
What it costs: a free account makes one video free, up to 0:45 of voice, with the Made with Kukarella.com mark on every frame. More videos come with a paid plan, where a one-minute video takes about 38 visuals (Kukarella's credits for pictures and video). A different Look or Shape means a new video, charged again.
Here is one I made from a listening dialogue for a class: a tired guest, a booking gone wrong, a hotel manager.
What it is and what it costs: Turn a Dialogue Into a Video: Make Video Explained. Every screen, step by step: How to Turn a Script Into a Video for Class.
Three quick questions before you start:
Frequently asked questions
Can Kukarella open my PowerPoint file?
No. Text to Voice takes a PDF, a Word file, plain text or pictures. Export the deck as a PDF, or copy the speaker notes, and bring that in. Each line of a PDF becomes its own paragraph, so join them into one paragraph per slide.
Is there a Kukarella add-in for PowerPoint or Google Slides?
No. You make the audio in Kukarella, download it, and insert the files into the deck yourself.
How long should each slide's narration be?
As long as the slide needs, but past about 100 words a slide usually holds two ideas: split it. At about 150 words a minute, 60 words take about 24 seconds.
Can different slides have different voices?
Not in one Text to Voice project: one voice reads the whole project, which keeps a deck consistent. For two speakers on a slide, write that part as a dialogue in DialoguesAI and download it separately.
Can I use the narration in a course I sell, or at work?
On a paid plan, yes: paid plans include commercial use. Audio made on the free plan has no commercial rights.
My AI voice sounds flat. What can I change?
Usually the voice and the script, before any setting: How to Make Text to Speech Sound Less Robotic.
Does Google Slides have its own AI voice?
Not in Slides itself. File > Convert to Video sends the deck to Google Vids, which can add an AI voiceover and read your speaker notes, on eligible paid Workspace and Google AI plans.
