Almost every transcription tool has a free tier, and almost every free tier exists to sell you the paid one. That is fine — it just means the useful question is not “which is free” but where does this one stop, because you will find out eventually and it is cheaper to find out now.
The short answer: if you are transcribing files you already have, the walls that matter are import counts and per-file length caps, and they arrive faster than the minute counters suggest. Tools that run on your own machine have no wall at all, because there is no per-minute cost for anyone to meter.
Where each free tier ends
| Tool | The wall | Good for |
|---|---|---|
| TranscriptSnap | Your device’s memory on very long files | Files you already have, any volume |
| Buzz | None — it is free software | Anything, if you will install an app |
| MacWhisper (free tier) | Larger models are Pro-only | Mac users trying local transcription |
| TurboScribe | 3 files/day, 30 min each | Evaluating the paid plan |
| HappyScribe | 10 minutes of AI transcription, ever | Meeting recording, not file imports |
| Otter.ai | 3 file imports, lifetime | Live meetings, not existing files |
| YouTube auto-captions | The video has to be on YouTube | Content that is going public anyway |
| Google Docs voice typing | Live speech only — cannot open a file | Dictating, not transcribing |
Two of those deserve expanding, because they are the ones people misread.
The Otter number people miss
Otter’s free tier advertises 300 transcription minutes a month, which sounds generous. The limit that actually binds is further down the pricing table: three lifetime audio or video file imports. Not three per month — three in total, for the life of the account.
This is not a trick; it follows from what Otter is. The product is built around joining your meetings and transcribing them live, and the free tier is generous about exactly that. Importing a recording you made elsewhere is a paid feature in all but three instances. If your job is “I have a folder of MP3s,” the free tier will stop you on the fourth one and no amount of unused monthly minutes will help.
Otter also caps a single free conversation at 30 minutes and keeps only your 25 most recent.
The HappyScribe number that is easy to miss
HappyScribe’s free plan advertises unlimited meeting recordings, and that part is true — its notetaker will join your calls and record them at 45 minutes a session, for nothing.
The number that matters if you have files already is elsewhere on the pricing table: a 10-minute trial of AI transcription, subtitling and translation. Not per month. Ten minutes, once. After that, uploaded files need a paid plan — $8.50 a month for 1,440 minutes a year, or top-up credits at $0.20 a minute.
Free video exports are also watermarked, which matters if you were planning to use it for subtitles.
This is the same shape as Otter: the free tier is generous about the thing the product is really for (live meetings) and close to zero on the thing people arrive from a search engine wanting (transcribing a file they already have).
The TurboScribe number people misjudge
Three files a day with a 30-minute ceiling. The daily reset makes it feel roomier than it is — you can transcribe 90 files a month, as long as you never need more than three on any one day and never have a recording longer than half an hour.
For evaluating the product this is plenty. For a week where you actually have work to do, it is not. The Unlimited plan is $10 a month billed annually, and if you need 10-hour files or speaker labels it is a fair price.
The tools with no wall
The pattern in the table is not a coincidence. Every tool with a hard free limit runs the model on its own servers, where each minute of audio costs real GPU time. Metering is not greed; it is the only way that shape of business survives contact with a heavy user.
Tools that run the model on your machine have the opposite problem — they have no per-minute cost to recover, so there is nothing to meter and no incentive to build a limit.
- Buzz is MIT-licensed and runs offline on macOS, Windows and Linux. Free in the software-freedom sense, not the free-tier sense.
- MacWhisper has a free tier that covers ordinary transcription; the larger models and batch features are in the one-time-purchase Pro version. Mac, iPhone and iPad only.
- TranscriptSnap — this site — runs Whisper compiled to WebAssembly inside the browser tab. No install, no account, no cap. Its wall, such as it is, is physical: your device does the work, so a phone is slower than a laptop and a multi-hour file can exhaust mobile memory.
The two free options that are not really transcription tools
Worth knowing about, because they are genuinely free and occasionally the right answer.
YouTube auto-captions. Upload a video, wait, download the generated captions. The quality is decent and the price is zero. The catch is structural: your video has to be on YouTube. You can set it unlisted, but the file is on Google’s servers either way, so this is only sensible for content that was going to be public.
Google Docs voice typing. Free and surprisingly accurate, but it transcribes a live microphone — there is no way to open an existing file. People sometimes try playing a recording into their laptop mic to get around this. It works badly: you get room noise, echo and a transcript noticeably worse than any tool that reads the file directly.
Picking by situation
- A folder of recordings to get through — a local tool. Every cloud free tier will stop you partway.
- A sensitive recording — a local tool, and for a different reason. See the private transcription comparison.
- Live meetings, and you do not mind the cloud — Otter’s free tier is well shaped for this and the 300 minutes go further than they sound.
- One long file, once — TurboScribe’s free tier if it is under 30 minutes; otherwise split it locally or pay for a month.
- A video that is going on YouTube anyway — let YouTube caption it and edit the result.
What “free” costs everywhere
One caveat that applies to all of these, paid tiers included. Speech models do not leave gaps when they are unsure — they produce the most plausible word instead, so errors read as confidently as correct text. Proper nouns, technical terms and overlapping speakers are where mistakes cluster. Budget an editing pass regardless of what you paid, and see why transcription tools invent words for what that failure actually looks like.