Privacy

What we process

When Listen is on, the sound from your microphone is sent to Deepgram, a speech-to-text service, so it can turn speech into captions. It is sent only while Listen is on. We don't keep the audio or the captions, and we ask Deepgram not to keep them either. The setting we use is called mip_opt_out, which opts out of Deepgram using audio to improve its models.

When you request beta access, we keep the details you type into the form (facility, your name, your role, your work email, how many staff would test, and any notes) so we can review the request and reply. They are also emailed to the team.

What stays in your browser

Saved conversations stay in your browser, on your computer. We never receive them. If you clear your browser's site data, they are gone. The only way a conversation reaches us is if you choose to include it in a problem report (see below).

What we count

We count how the app is used so we can learn whether it helps and where it fails. Each count is only a number, a yes or no, or a pick from a short fixed list. None of them can carry words. In plain terms, we count when:

• the app is opened;
• a conversation starts;
• you listen: how many seconds of captions, which caption service was used, and whether the signer or the face-to-face view was on;
• you reply, and whether you typed it or tapped a phrase;
• the split view is turned on or off;
• the captions run into an error, and the kind of error;
• the app itself runs into an error;
• you report a problem, and only the category you picked.

We never record what was said, what you typed, or which phrase you tapped. Our server only accepts counts from this fixed list and throws away anything else. Counts are linked to your facility and your sign-in, so we can tell whether people come back.

When you report a problem

The Report a problem button is the one place the app sends something you write. You choose what goes in it. You write what happened and pick a category. If you tick the box to include the conversation, you see exactly which lines will be sent and can remove any of them before you send. Nothing is sent until you press send.

Each report also carries your facility, your sign-in email, your browser, the app version, which caption service was in use, and the time. It never includes audio or video.

The report is emailed to the team. We don't store its text in our database, and our own code doesn't write it to logs. It stays in the team's email inbox like any other email, and our email service (Resend) keeps a copy of sent emails for a limited time.

Who sees it

The See The Speech team. Access requests and problem reports arrive in the team's inbox. Usage counts are read as totals for each facility in a weekly summary email, which never contains any words from a conversation. The services below handle information on our behalf to run the site. Two of them also keep copies for a limited time: Vercel keeps basic request logs, and Resend keeps copies of sent emails.

The services we use

• Vercel hosts this website and runs its server code. It keeps basic request logs, such as your IP address and browser, for a limited time.
• Clerk handles sign-in with an email code, and your facility's list of testers. It sets the cookies needed to keep you signed in.
• Neon is the database for access requests and usage counts.
• Resend sends our emails: access requests, problem reports and the weekly summary. It keeps a copy of sent emails for a limited time.
• Deepgram turns speech into captions while Listen is on.

How to ask for deletion

If you asked for beta access, reply to the email we send you about your request and say what you would like removed. We'll tell you when it is done.

If you are a tester, use the Report a problem button in the app and say what you would like removed (your sign-in or your usage counts, for example).