Real-time transcription without the cloud, for medical interpreters
More and more interpreters use real-time transcription so they do not miss a number or a drug name. The question almost nobody asks is where that audio is processed: on another company's server or on your own computer.
In a medical call, what is said is protected health information. This page explains the difference, what the HHS guidance says about the cloud and how to check for yourself what a tool does. It is not legal advice: what binds you is your contract and your agency's policies.
Cloud or local: what changes
- Where the audio goes
- In the cloud, the call audio travels to the provider's servers to be transcribed. Locally, it is processed on your computer and never leaves it.
- What is kept
- Many cloud tools keep audio or text for history, support or to improve their models. A local tool may keep nothing, but it is worth checking.
- Who else handles it
- In the cloud another company is involved, with its staff, vendors and backups. Locally, only your computer.
- Offline
- A local tool keeps working if the connection drops; a cloud tool does not.
- What the cloud gives you
- Bigger models and access from any device. What local gives you: the audio never leaves your computer.
What HHS says about the cloud
According to the U.S. Department of Health and Human Services (HHS) guidance on HIPAA and cloud computing, a cloud service provider that creates, receives, maintains or transmits electronic protected health information on behalf of another is a business associate and needs a business associate agreement (BAA). That applies even if the provider says the information is encrypted.
So if a cloud tool processes the audio of your calls, the question is not only whether it is good, but whether that agreement exists and who signed it. With a tool that processes everything on your computer, the audio does not reach any third party.
HHS guidance: https://www.hhs.gov/hipaa/for-professionals/special-topics/health-information-technology/cloud-computing/index.html
How to check whether a tool is local
- Offline test
- Disconnect wifi or the cable and try it. If transcription stops, the audio is going to a server.
- Account
- If it asks you to sign in to transcribe, there is almost always a server behind it.
- Privacy policy
- Look for mentions of uploading, storing or using audio or transcripts to train models.
- Files
- Check whether it saves transcripts or recordings on your disk, and where.
- Network activity
- Windows Resource Monitor (Network tab) shows which programs send data while you talk.
How Terpside does it
Terpside uses a local speech recognition engine (Whisper, by OpenAI, MIT license) and a local translator (Opus-MT, by the University of Helsinki). Both are installed with the program and run on your Windows computer: nothing from the call leaves it, and it works offline.
It listens only to the call's audio output (the microphone is never opened), has no account and no save button, keeps the transcript in memory while the program is open and discards it when you close it, and the window shows up black in screenshots and screen sharing. Its only outbound connection renews a paid license, with no call data, and you can turn it off.
The honest limit: on a modest computer, a local model is smaller than cloud models and may make more mistakes or run slower. That is why Terpside measures your computer at startup and uses the graphics card when it helps.
Frequently asked questions
Is a local tool automatically HIPAA compliant?
No. HIPAA obligations are yours and your agency's: protecting information, not sharing it and following their policies. A local tool lowers the risk because the audio never leaves your computer, but it does not replace your contract or your agency's policies.
Does Terpside work offline?
Yes. Recognition and translation run on your computer and the license is checked without a network. Only renewing a paid license uses the internet, and you can turn it off.
Is it as accurate as a cloud tool?
It depends on your computer. Cloud models are bigger; a local one on a modest computer may make more mistakes. In exchange, the call audio never leaves your computer. The best way to know is to try it on your own calls.
What about Windows or Chrome live captions?
According to Microsoft and Google, they are also processed on the device, but they were not made for interpreting: they rewrite the sentence while people talk, do not separate who said what and do not show a translation next to it.
Try it on your calls
Free Pro trial: up to 2 hours a day for 5 days, no account or card. After that it stays on Free with 60 minutes a day.