Is it safe to upload audio to a transcription tool?
If you’re transcribing a confidential interview, a client call or a therapy session, “just upload it” deserves a second thought. Your audio can be personal, privileged or commercially sensitive. Here’s what actually happens to a file when you transcribe it, and the questions worth asking before you trust any tool with it.
Where your audio goes
Most AI transcription tools don’t run the model themselves - they send your audio to a speech-to-text provider (like OpenAI Whisper) for processing, then hand you the text. So there are really two questions: who receives the audio, and how long anyone keeps it.
The risky pattern is a tool that stores your recordings indefinitely on its own servers “so you can access them later” - because now your sensitive audio lives in a second place, under someone else’s retention policy, for as long as that company exists.
What Speecho does
Speecho is built to hold as little as possible:
- Audio is streamed for inference only. Your file goes to the AI provider to be transcribed and is not stored on our servers afterwards.
- Transcripts auto-delete after 90 days. The text output isn’t kept forever either - it’s removed on a rolling basis.
- No bot in your calls. Because you record locally and upload, your meeting audio never has to flow through a third-party notetaker bot’s account.
- Guest-first. You can transcribe without even creating an account, so there’s no profile accumulating your history.
Our full privacy policy spells out the details in plain language.
Questions to ask any transcription tool
Before you upload something sensitive - to us or anyone - get clear answers to these:
- Is my audio stored after transcription, or only processed? “Processed and discarded” is what you want.
- How long are transcripts retained, and can I delete them?
- Who is the audio shared with? A reputable tool will name its AI provider.
- Is my recording used to train models? For sensitive content, you want a firm no.
- Do I have to create an account? Guest use means less data tied to you.
If a tool can’t answer these clearly, that itself is the answer.
Practical habits for sensitive recordings
- Get consent when you record other people - it’s often legally required, and always the right thing.
- Strip what you don’t need. If only part of a call is relevant, trim before uploading.
- Use speaker labels instead of names where you can, and keep identifying details out of file names.
Sensitivity is exactly why the pay-as-you-go, no-bot, nothing-stored model exists. If you want the details first, read our privacy policy - or transcribe a file free and see the workflow for yourself.
