Private audio transcription for clinical notes
You dictate a letter after a clinic. You record a supervision session. You capture a phone consultation for the file. The audio contains patient names, NHS numbers, clinical findings and professional opinions. Where that audio goes for transcription is a data governance question, and for most tools the answer is a US cloud.
This is a walkthrough of how Hush Echo handles audio transcription: GPU-accelerated speech-to-text on hardware we own, with no audio retained after the transcript is returned.
The workflow
Open hush-ai.uk/echo. You can record directly in the browser or upload a file. Hush Echo supports WAV, MP3, MP4, M4A, FLAC, OGG, WEBM, AAC and WMA. There is nothing to install.
The audio is transmitted over TLS to hardware we own in the UK and processed using GPU-accelerated speech-to-text. The transcript is returned to your browser.
What happens to the audio: processed in memory on our GPUs, then gone. The audio file is not written to disk on our servers, not stored in any database, not queued for model training. There is no third-party transcription API in the chain. The infrastructure is the same hardware that runs the rest of Hush, owned and operated by us, outside US CLOUD Act jurisdiction.
Read the transcript, correct any misheard words (medical terminology, proper names), and copy it where it needs to go: into a clinical letter, a file note, a referral, or your clinical system. If you want Hush to draft a document from the transcript, you can do that in the same session: "Turn this transcript into a structured discharge summary" or "Extract the action items from this meeting."
The audio is gone from our servers after transcription. The transcript text is available in your browser session and, if you are signed in, stored encrypted on your device as part of your chat history. You can delete it in one click. The audit log records that a transcription took place, not the content.
Why the infrastructure matters for audio
Audio is harder to redact than text. A patient's name spoken aloud is embedded in the waveform; you cannot strip it the way you might remove a name from a text field before submitting. This makes the question of where the audio goes more important, not less.
Consumer transcription services route audio through cloud APIs operated by US companies. Even where the terms say the audio is not used for training, the audio has still been transmitted to and processed by infrastructure under US jurisdiction, subject to legal process under the CLOUD Act. For a dictated clinic letter that names patients, that is a fact your DPIA should address.
Hush Echo processes audio on GPUs we own. The chain is: your browser, TLS, our hardware, transcript back, audio gone. No third-party. No US entity. If that sounds like a claim you should check rather than trust, good: the verify-us page has the terminal commands.
Use cases beyond clinical dictation
- Supervision and CPD: record a supervision session, transcribe it, and produce a structured supervision note. The recording contains clinical detail about patients discussed; it should not go through a US cloud transcription service.
- Legal dictation: solicitors dictating attendance notes, file notes and instructions. Client-privileged material, same jurisdiction and retention concerns.
- HR and investigations: disciplinary hearing recordings, grievance interviews. Special-category employee data in audio form.
- Meetings: any meeting where sensitive commercial or personal information is discussed. The transcript is useful; the audio going to a third-party cloud is not.
Try a transcription
Record a short clip or upload a test file (not patient audio, use something innocuous first). See the transcript quality and verify the privacy claims before putting real data through it.
Open Hush Echo →Echo is included in all Hush plans. The free tier lets you test the transcription without an account.
About Hush AI: Hush AI (hush-ai.uk) is a private AI platform built by its founder, Dr W.J Carter. All processing, including audio transcription, runs on hardware we own in the UK. No audio or text is stored on our servers, and nothing is used for training. The company has no US parent and no US cloud sub-processor.