In Thing 9, you turned words into pictures. In this Thing, you're going to turn words into sound of a different kind.
AI can now clean up a recording so your kitchen sounds like a studio, and let you edit audio simply by editing the words on screen. Delete a word from the transcript and it disappears from the recording. And then there's music. Type a short description (a genre, a mood, perhaps some lyrics) and AI will give you back a complete song, with vocals.
That's exciting and a little unsettling in roughly equal measure. These tools solve real problems, like tidying up a voice recording or finding background music for a presentation. They also raise real questions about creativity, ownership and copyright. In the activity, you'll edit a short recording of your own and create a short piece of music, and decide for yourself how convincing the results are.
Listen closely as you go. Learning to hear the difference between real and synthetic sound is a skill that will matter more and more.
Editing audio by editing text
When the transcript becomes the editing tool, cutting audio feels like tidying a document.
Traditional audio editing means staring at a waveform, finding the right spot by eye and hoping you don't chop the start off a word. It works, but it's fiddly and feels disconnected from what was actually said.
AI audio editing flips this round. The tool transcribes your recording, using the speech-to-text technology you met in Thing 8, and links every word on screen to the moment it was spoken. Delete an "um" from the transcript and the matching audio goes with it. Cut a rambling paragraph and it's gone from the recording. You make decisions at the speed of reading rather than the speed of listening.
Descript
Descript popularised this idea and is still the reference point for text-based editing. It looks and feels like a word processor. You can remove filler words one at a time or in bulk, cut whole sections and rearrange your recording by moving text around.
Descript runs in your web browser (Chrome or Edge work best), with a desktop app if you prefer. The free plan gives you 60 media minutes a month, plenty for a short recording. AI extras such as Studio Sound (its own clean-up tool) and automatic filler word removal draw on a small one-off allowance of AI credits, so use them deliberately.
Where it falls short
Listen carefully to the joins after an edit. Sometimes the cut is seamless. Sometimes there's a tiny click, an unnatural pause or a sudden change in background sound, especially if your pace changed between the words either side. A steady speaking pace gives the cleanest results. And if the transcript mishears a word, check the audio before you cut around it.
AI audio clean-up
You no longer need a quiet room and an expensive microphone to sound clear.
Editing changes what was said. Clean-up, sometimes called enhancement, changes how it sounds. This used to need specialist skills. Now it's one step: upload your recording and listen to the result.
These tools have learned to tell human speech apart from everything else: traffic, a humming fridge, keyboard clicks, room echo, the dog next door. They isolate the voice, reduce the rest and make the voice clearer. A phone recording made at your kitchen table can come out sounding like it was made in a small studio.
Adobe Podcast Enhance Speech
Adobe Podcast Enhance Speech does one thing well, with no settings to learn. Sign in with a free Adobe account, upload a recording and download the cleaned-up version. The free tier takes audio files up to 30 minutes long, with up to an hour of processing a day. You may see it described as part of Adobe Firefly, as Adobe is folding its podcast tools into its wider AI range. It's designed for speech, not music.
Clipchamp
If you use Windows, you may already have an AI clean-up tool. Microsoft Clipchamp is a video editor that comes with Windows 11 and also runs in a browser. The free version includes AI noise suppression, an auto cut feature that removes pauses and automatic captions. Many organisations have the work version through Microsoft 365, with the same audio features.
Listen for over-processing
Clean-up tools can overdo it. When there's a lot of noise to remove, voices can come out slightly watery or metallic. The harder the tool works, the more likely it is to add odd sounds of its own. And if the background noise is louder than your voice, even AI can only do so much.
AI music generation
You describe the song. The AI writes, sings and plays it, usually in under a couple of minutes.
You write a short description of the music you want (genre, mood, tempo, perhaps your own lyrics) and the AI generates a finished-sounding song with verses, a chorus and convincing human-sounding vocals.
These tools are trained on vast amounts of existing music. They've learned patterns, like how a pop chorus builds or how a reggae drum pattern differs from dance music. They don't stitch together pre-made clips. They generate new audio guided by those patterns, which is why the question of what they were trained on matters so much.
What you can control, and what you can't
You have most influence over:
- genre and style, such as "indie folk" or "orchestral film score"
- mood and energy, such as "upbeat", "melancholic" or "anthemic"
- instruments, such as "acoustic guitar and strings", though the AI makes the final choices
- lyrics, which you can write yourself or leave to the AI
- whether there are vocals at all, if you want an instrumental background track
What you can't easily do on a free plan is fine-tune the result, like "make the second verse quieter". If you don't like something, change your description and generate again. It's less like using a recording studio and more like briefing a very fast session musician.
Listen closely to AI-written lyrics. They rhyme and scan, but lean on familiar phrases like "rising up" and "breaking free". Writing your own, even simple ones, usually makes a more interesting song because it puts your ideas at the centre.
Suno
Suno is the most popular AI music generator and the best place to start. Give it a simple description, or switch to custom mode to add your own lyrics and a style description. Each request gives you two versions to choose between.
The free plan gives you 50 credits a day, enough for roughly ten songs, and credits don't roll over. Free songs are for personal, non-commercial use only. Since September 2026, free accounts get just seven downloads in total, ever, but every song stays playable and shareable by link, so save share links rather than downloading. Paid plans add commercial rights, monthly downloads and more editing tools.
Gemini
Google's Gemini app can make music too, using Google's Lyria model. Choose the create music option, describe what you want and Gemini makes a track of around 30 seconds with its own cover art. Type "Lyrics:" before any lines you want sung. It's free for users aged 18 and over, with higher limits on Google's paid AI plans.
Every Gemini track carries SynthID, Google's invisible watermark that marks it as AI-generated, and you can upload an audio file to Gemini and ask it to check for that watermark. The tracks are much shorter than Suno's, but it's a quick way to hear how a second tool interprets the same brief.
Other tools worth knowing about
Google Flow Music began life as Riffusion, one of the first AI music tools, and is now part of Google Labs. You chat with it to build a full-length song, then remix or extend sections. It also uses Lyria and SynthID. Check whether it's available to you, and what's free, when you visit. You may also come across Udio, Suno's main early rival, which has switched off downloads on all plans while it builds a licensed service.
What these tools are good for
Most of the time, "good enough" is exactly what you need, and these tools get you there fast.
Put the wow factor to one side and there are plenty of everyday uses:
- tidying up a voice note or recorded talk before you share it
- cutting the stumbles and long pauses out of a recorded presentation
- rescuing a recording made somewhere noisy, like a busy café
- making a short jingle for a podcast or a community group's video
- creating background music for a slideshow or presentation, instead of stock music everyone recognises
- writing something silly for a friend's birthday
None of these tools will turn a truly terrible recording into broadcast quality, and AI music won't replace a skilled composer who understands what a project needs. But they can take something you'd be embarrassed to share and make it presentable, quickly.
It's a pattern you'll see throughout the programme. AI keeps removing technical barriers between having an idea and acting on it, just as you didn't need design skills to make images in Thing 9. What matters more are the editorial skills: knowing what you want to say, recognising what sounds good and choosing your tools thoughtfully.
Voices, creativity and copyright
AI music is built on human music, and the people who made it are still arguing about the bill.
Whose voice is that?
AI singers can sound remarkably real. In a survey for the streaming service Deezer in late 2025, 97% of people couldn't tell AI music from human-made music in a blind test. By mid-2026, more than half the new tracks uploaded to Deezer each day were AI-generated, though they made up only a small share of listening. Deezer now labels AI-generated albums, and Spotify only allows music that imitates an artist's voice when that artist has given permission.
Suno says it strips real artists' names out of prompts and works from a description of the style instead. Not everyone is convinced that's enough. In September 2026, a group of musicians sued Suno, claiming its model had captured distinctive features of their voices without consent. Voice cloning is covered in Thing 8 and voice scams in Thing 16. The point for now: a convincing voice is no longer proof that a person sang or said something.
Copyright and the music industry
The models behind Suno, Udio and similar tools were trained on enormous amounts of existing music. The music industry's position, backed by the major record labels and many artists, is that this was copyright infringement on a massive scale. In 2024, major labels sued both Suno and Udio.
The legal picture has shifted since then. In late 2025, Universal Music and Warner Music both settled with Udio, and Warner settled with Suno, in each case swapping lawsuits for licensing partnerships. Suno's latest models were trained partly on licensed Warner music. But the dispute is far from over. Universal and Sony are still suing Suno, and Sony is still suing Udio. In July 2026, a German court ruled that Suno can't train on copyrighted songs without a licence (a ruling that isn't yet final). The direction of travel looks like licensing rather than outright opposition, but the law is still being settled.
In the UK, the government dropped its preferred plan to let AI companies train on copyrighted work unless creators opted out. In March 2026, it said it would gather more evidence and watch how licensing develops before deciding what to do. Some AI companies have chosen to train only on licensed material, and an independent scheme called Fairly Trained certifies those that do.
Licensing deals between big companies don't settle the ethical questions. Many individual musicians feel their work was used without consent to build systems that now compete with them, and reasonable people disagree about whether AI music devalues human creativity. Thing 14 looks at fairness in more depth.
What this means for you
Songs made on free plans are for personal, non-commercial use. You can make them, share them with friends and submit them for your badge, but don't use them for anything commercial, such as an advert. That needs a paid plan with the right licence. Be open about AI generation if you share your creations publicly, because passing off an AI track as your own performance would be misleading. And remember the download caps: Suno's free accounts get only seven downloads in total, so treat share links as the normal way to keep your songs.
Resources to explore
The reference tool for editing audio and video by editing a transcript. Works in your browser, and the free plan includes 60 media minutes a month.
One-click AI clean-up for speech recordings. Free with an Adobe account, for files up to 30 minutes and an hour of processing a day.
The most popular AI music generator, with a free plan of 50 credits a day for personal, non-commercial use. Free accounts can share songs by link but only download seven in total.
Google's own short guide to making music in Gemini, including how to add your own lyrics and steer the style, vocals and tempo.
The independent certification scheme for AI companies that train on licensed rather than scraped content. Useful background for the copyright questions in this Thing.
The UK government's March 2026 report explaining why it stepped back from a broad copyright exception for AI training and what it plans to do next.
Activity: your voice, your track
This activity has two parts. In part 1, you'll record yourself talking and use AI to edit and clean up the recording. In part 2, you'll create a short song or background track and compare two music tools.
What you'll need: a computer with a microphone (built-in is fine) or a phone to record on, a free Descript account, a free Adobe account, a free Suno account and, for the comparison, a Google account (Gemini music is for users aged 18 and over).
Part 1: edit and clean up your own recording
- Record a short clip. Record yourself talking for about a minute about a hobby, a place you'd recommend or something you're learning. Don't script it. A few "ums", restarts and some background noise give the AI something to work on. Use your phone's voice recorder or your computer's microphone, and keep a copy of the original.
- Edit it by text in Descript. Go to descript.com, create a free account and work in your browser (Chrome or Edge work best). Start a project and upload your recording. When the transcript appears, remove a few filler words, then delete a whole sentence from the middle. Play it back, listen carefully to the joins and export the edited audio.
- Clean it up with Adobe Podcast. Go to podcast.adobe.com/enhance, sign in with a free Adobe account and upload your original recording (not the Descript version). Compare the enhanced version with the original, then download it. No Adobe account? Try Descript's Studio Sound if you have credits left, or the noise suppression in Clipchamp.
Part 2: create a song or background track
- Make a track in Suno. Go to suno.com and create a free account. Choose one of these:
- a short song: switch to custom mode and write a verse and a chorus about the hobby or place from your recording, using tags like [Verse] and [Chorus], then add a style description such as "upbeat indie folk, acoustic guitar and handclaps, warm and nostalgic"
- a background track: describe an instrumental piece that could sit under your recording, such as "calm lo-fi piano, gentle and unhurried", and switch on the instrumental option
Be specific, because vague prompts like "a nice song" give bland results. Listen to both versions, save your favourite and copy its share link rather than downloading (free accounts get only seven downloads in total). Note down your prompt and lyrics.
- Compare it with Gemini. Open gemini.google.com, choose the create music option and give it the same brief. If you wrote lyrics, type "Lyrics:" before a short section of them, as the tracks are only around 30 seconds. Copy the share link. Can't use Gemini? Make a second Suno track with the same lyrics in a completely different style.
- Reflect. Listen back to everything you've made and write 250 to 350 words covering:
- what the text editing changed about your recording, and whether it lost any natural character
- what the clean-up changed about the sound, and whether you heard any odd artefacts
- how the two music tools interpreted the same brief, and how AI-written lyrics (if Suno added any) compare with your own
- where you could genuinely use these tools in your own life
- your view on the voice and copyright questions in this Thing
Your output
Your original, edited and enhanced recordings (or screenshots of your edits if you'd rather not share your voice), your written reflection and share links to your Suno and Gemini tracks with the prompts and lyrics you used.
Claim your Open Badge
Submit your three versions of your recording (or screenshots of your edits), your 250 to 350 word reflection and the share links to your Suno and Gemini tracks with your prompts and lyrics via cred.scot.
Submit your edited audio, your music share links and your reflection to claim this badge via cred.scot.
Claim now