Where the cost of transcription normally goes
Cloud transcription runs on rented servers with expensive graphics chips. Every minute of audio you upload occupies one of those servers for a while, so each minute has a real cost the company has to recover. That is why most services sell minutes, cap free plans and push you toward a subscription.
Transcribela does the work somewhere else: on your own device. The open-source Whisper speech model is sent to your browser once, and your computer or phone runs it. Our server only delivers web pages, which costs the same whether you transcribe one minute or a hundred hours. That is the whole reason there is no limit.
What runs where
| Part | Where it runs | Cost per extra transcript |
|---|---|---|
| Speech recognition | Your device, inside the browser tab | None to us; a little battery for you |
| Speech model download | Hugging Face’s public network, once per browser | None; 64 MB to 760 MB once, depending on the setting |
| Pages, scripts and styles | Our web host and its cache | Effectively none |
| Your recording and transcript | Your device only | Nothing is stored by us |
How the site pays for itself
Through advertising on content pages, shown by Google AdSense. Ads are labelled, kept away from the transcription controls and never appear inside your transcript or your exports. Visitors in the European Economic Area, the UK and Switzerland are asked for consent first, and anyone can change their choice on the privacy choices page.
Because each extra visitor costs almost nothing, a modest amount of advertising covers the site even with a very large number of users. There is no investor plan to switch on charges later; the design simply does not need them.
What we do not do
- We do not upload, store or listen to your recordings. We never receive them.
- We do not sell personal information.
- We do not limit minutes, files or exports to push you to a paid plan. There is no paid plan.
- We do not train models on your transcripts, because we never have them.
- We do not ask for an account, an email address or a card.
The honest trade-offs
Free and private comes with a few differences from paid cloud services, and it is better to know them up front.
- The first run downloads a model. The app shows the size and asks first. Fast is 64 MB and Balanced 136 MB; the larger settings are worth it on Wi-Fi if you transcribe often.
- Speed depends on your device. A recent laptop runs much faster than real time on Balanced; an older phone is slower. Paid services use data-centre hardware, so they can be faster on old devices.
- No automatic speaker detection. You label speakers by hand, which takes a minute for a typical interview.
- No cloud sync or team workspace. Your library lives in your browser. Use backup and restore to move it between devices.
For confidential recordings, the trade-off usually runs the other way: a service that never receives your audio is simpler to approve than one that stores it.
Will it stay free?
Yes. The costs that force other services to charge do not exist here, and every feature on the site today is free. If we ever add something that genuinely costs money to run, such as optional cloud features, it will be separate and optional, and everything that works today will keep working at no cost.