Transcribe voice memos to text
Four voice memos as readable text, word for word, without the audio leaving the computer.
A real run on 13 November 2026, on a Linux computer, using Claude Code with Opus, on a Claude plan. The waiting is sped up; it took 3 min 50 s in real time.
The short answer.
On a Mac with macOS 15 or an iPhone with iOS 18, Voice Memos shows a transcript for each memo that you can copy out, one at a time. Microsoft Word for the web can transcribe uploaded recordings with Microsoft 365. Online services like Otter work but upload the audio and cap free minutes. Stella (stella.sh) transcribed a folder of memos with a free speech tool it installed on the computer, so the audio wasn't sent to a transcription service.
Stella does this on Windows, Mac and Linux.
What happened.
Four transcripts in about 3 minutes 50 seconds, including installing the speech tool. Checked afterwards against the scripts the memos were recorded from: three transcripts word for word (0% word error rate) and the fourth at 3.6%, with every key name and fact present, and the recordings unchanged.
The memos were voiced from scripts written for this run, with a light room tone; the people and places are fictitious. Our check script needed a fix to run on Linux and was re-scored after the run, with the answer key unchanged. The work is sped up 15×.
How Stella does it.
- Finds the four voice memos in Downloads, including a two-minute lecture
- On the free plan, installs a free speech-recognition tool that runs on the computer itself (about 2 GB) instead of a paid transcription service
- Writes one text file per memo, named after it, in a new Voice memo transcripts folder
- Makes one page with every transcript and a one-line summary of each at the top, and lists what each memo says in the chat
- Says that names and place names are best guesses, and offers to keep the speech tool for next time or delete it
The prompt.
Turn the voice memos in my Downloads into text I can read.
This is the exact prompt from the recording. If your files are somewhere else, say where.
Do it yourself.
Mac or iPhone: Voice Memos
On macOS 15 or iOS 18 and later, open a memo in Voice Memos and click or tap the transcript button (a speech bubble) to see the text. Select it and copy it into Notes or a document. It works one memo at a time, in English and a few other languages.
Windows or the web: Word's Transcribe
With Microsoft 365, open Word for the web, choose Home, Dictate, Transcribe, then Upload audio. It returns a transcript with speakers marked that you can add to the document. The audio goes to Microsoft, and there's a monthly limit on uploaded minutes.
Free and local: Whisper
OpenAI's open-source Whisper model runs on your own computer through free apps and tools (whisper.cpp, faster-whisper, or desktop apps built on them). The bigger models are more accurate but download a gigabyte or more and run slowly without a good graphics card.
Online services
Otter, TurboScribe and similar sites are quick and add speaker labels, but they upload your recordings and cap the free minutes. Fine for a public lecture; think twice for private notes.
Fix the names
Speech recognition guesses at names, places and jargon. Skim each transcript for capitalised words and correct them; most tools let you add a vocabulary list or prompt to get them right next time.
Where it stops.
- Names and unusual words are best guesses. Stella said so; tell it the right spellings and it can correct them.
- The free local tool is slower than a hosted service and takes about 2 GB of disk space. Stella offered to delete it afterwards.
- The audio stayed on the computer, but the finished transcripts go to the AI model you use in Stella when it summarises them. Ask for transcripts only if you'd rather it didn't read them.
- The test memos were clear speech with light background noise. Noisy rooms, several people talking at once or strong accents will be less accurate. Labelling who is speaking wasn't part of this test.
- Recorded on Linux. On a Mac or Windows PC Stella installs a similar local tool, which wasn't recorded.
Questions.
Can I transcribe voice memos for free?
Yes. Voice Memos on macOS 15 and iOS 18 shows transcripts for free, one memo at a time. Whisper-based tools run free on your computer, and Stella used one for a whole folder.
Does my audio get uploaded?
Not in the recording: the speech tool ran on the computer. Word's Transcribe and online services do upload it. The text Stella then summarised does go to the AI model you use in Stella.
How accurate is it?
On these clear test memos, three came out word for word and the fourth at 96% of words right. Expect more mistakes on phone calls, noisy rooms and crosstalk.
Can I transcribe a video or a long lecture?
Yes, the same way; longer audio just takes longer locally. To put the words on a video as captions, see burn subtitles into a video. For music files, see edit MP3 tags in bulk.
Is Stella free for this?
Yes. In the recording the account was on Stella's free plan, so Stella installed a free speech tool on the computer and nothing was paid; it's just slower. Stella's own built-in transcription isn't part of the free plan.
Can I control what Stella changes on my computer?
Yes. Stella works on Windows, Mac and Linux with your own files and shows each step in the chat, so you can watch it and stop it at any point. Tell it what it may not do, as the prompts on these pages do (“don't delete anything without asking”). In the Downloads recording it deleted nothing until it was told to go ahead. It also refuses commands that would wipe a whole disk, your home folder or system folders.
- Apple: View a transcript in Voice Memos on Mac
- Microsoft: Transcribe your recordings
- OpenAI Whisper on GitHub
Checked against these sources on 13 November 2026.