Audio and video
AI can write down what people say, read text aloud in a voice, and make short videos. Think of a note-taker, a voice actor and a tiny film studio. Learn which one you need, how to use it, and when to be careful.
First, what do you think?
You get a voice message from your son. He sounds upset and asks for money. Could AI have made that voice? Have a guess.
What can AI do with sound and video?
The whole story in three sentences. AI can turn speech into text, and text into speech. It can also make short videos from a description. Each one is a predictor, trained on a huge amount of sound or video.
Think of three helpers. A note-taker writes down what people say. A voice actor reads your text aloud. A tiny film studio makes a short clip from your description.
- Speech to text. Subtitles, notes of a meeting, a voice memo as text. One model learned from 680,000 hours of audio. It came close to how well people type out speech.
- Text to speech. A story read aloud, a voice-over for a presentation. Tools like ElevenLabs offer this, and also let you design a voice.
- Text to video. A short clip from a description, with sound. Google's model makes these, with an invisible watermark.
A voice can also be copied. The Fraudehelpdesk calls a copied voice or face a deepfake. Scammers use copied voices to ask for money.
What the predictor does
Each helper predicts what fits. That's the rule.
The note-taker predicts which words were said. Usually it's right. But names, mumbling and background noise make it guess. "Close to people" isn't perfect.
The voice actor predicts how your text should sound. It doesn't understand what it reads. A copied voice predicts how that person would sound. It can say things the person never said.
The film studio predicts a clip that fits your description. Like with images, it shows what fits the words, not what happened.
Speech to text, text to speech, text to video: three predictors. Each one makes something that fits. Not something that's guaranteed to be true or real.
Try it yourself
Which helper do you need? Sort each task.
Everything runs in your browser. Nothing you type is sent anywhere.
How do you use it well?
Four steps. Start today with a voice memo: say it, and let your phone write it down.
Pick the right helper
Do you have speech and want text? Text and want a voice? A description and want a clip? Pick the tool that does that.
Check the transcript
Read what the note-taker wrote. Fix names, numbers and words it misheard. Don't quote anyone from a transcript you haven't checked.
Describe the clip
For a video, describe the scene, how the camera moves, the light and the mood. Say what sound you want. To change one thing, add: "Keep everything else the same." That's Google's own advice.
Ask first, and say it's AI
Only use someone's voice or face if they agree. Say when a voice or video is made with AI. In the EU, deepfakes must be labelled.
In practice
Three people meet the three helpers.
Case 1The interview transcript
Fatima (16) interviews her grandmother for a history project. She lets an app turn the recording into text. Her grandmother's village is spelled wrong, and one date is off.
Copy the transcript straight into the project.
The note-taker predicts words. Names and numbers are where it guesses.
Listen back and check names, places and dates. Then use the corrected text.
The app saves hours of typing. Fatima makes sure it's right.
Case 2The voice message
Kees (60) gets a voice message. It sounds like his son, who says he lost his phone and needs money fast.
Pay. It's his son's voice, after all.
The Fraudehelpdesk warns about exactly this. A "son" or "daughter" asks for money, sometimes with a copied voice.
Hang up. Call his son on the number he already has. Never pay from a message alone.
A copied voice can't answer the real phone.
Case 3The bakery clip
Ilse (33) runs a small bakery. She wants a short clip for social media. Her first try, "bread", gives a dull video.
Try "bread" again, or give up.
"Bread" leaves everything open.
Describe it. Warm morning light. A slow camera move over fresh loaves. The sound of a busy street. Then post it as made with AI.
The clip fits her shop. And her customers know it's not a real film of her bakery.
When not to
- Don't copy someone's voice without permissionA voice can be used to deceive.Ask first, or use a made-up voice.
- Don't trust a voice that asks for moneyScammers copy voices.Call back on a number you already know.
- Don't quote from an unchecked transcriptNames and numbers get misheard.Listen back and fix it first.
By situation
Subtitles and transcripts help with interviews and presentations. Check names and dates. Some voice tools are 18+: ElevenLabs is one of them. Check the age limit first.
A transcript of a meeting saves time. Ask everyone if recording is okay, and check the text before you share it. Free plans may limit how you use the result. ElevenLabs allows free use only for non-commercial purposes.
A story read aloud, a voice memo as text, a short clip for a party. And one rule for the family. Anyone asking for money by voice message gets a call back first.
Myths and mistakes
| Claim | True? | Why |
|---|---|---|
| I'd always recognise my own child's voice | Not always | Voices can be copied. Call back to be sure. |
| A transcript is word for word correct | No | It comes close, but names and numbers get misheard. |
| An AI video shows what happened | No | It shows what fits the description. |
| Copying a famous voice for fun is harmless | No | It can deceive people. Ask, or use a made-up voice. |
Related topics
Read first
- Imagesthe same idea for pictures: describe, then improve.
Next
- Search and researchAI that searches the web for you.
Related
- Deepfakes and scamshow to spot and stop fake voices and videos.
Sources
- researchRadford e.a. (2022). Robust Speech Recognition via Large-Scale Weak Supervision.A speech recognition model trained on 680,000 hours of audio from the internet comes close to how accurately people write down speech.
- documentationElevenLabs. ElevenLabs Pricing.The free plan includes text to speech, speech to text, sound effects and voice design, with a monthly limit. (Checked through a search engine.)
- documentationElevenLabs. ElevenLabs Terms of Service.You must be at least 18. Free users may only use it for non-commercial purposes. (Checked through a search engine.)
- documentationGoogle AI for Developers. Gemini Omni Flash.Makes short videos with sound from a description or an image. Describe the scene, camera movement, lighting and mood. For edits, add 'keep everything else the same'. All videos carry an invisible SynthID watermark.
- organisationFraudehelpdesk. Fraudevormen en -termen.A deepfake copies someone's voice or face with AI and is hard to tell from the real thing. In 'help request fraud' a 'son' or 'daughter' asks for money by app, sometimes with a recorded copy of their voice.
- governmentEuropean Commission. AI Act.The EU AI Act sorts AI into four risk levels. Some uses are banned, like social scoring. People must be told when they talk to a chatbot, and deepfakes must be labelled. AI literacy rules apply from 2 February 2025.
Test
You get 7 questions about the whole page. Get 5 right, and you've mastered the topic.
About the tools in this course. Every tool is described in the same way, in alphabetical order, without a recommendation. Names, prices and features change often: the maker's own site is always right. This site was built with the help of Claude, one of the tools described here.