Status
Is Transcribe Today working?
Every five minutes our own server signs a grant, opens a WebSocket to stream.transcribe.today
and streams five seconds of speech, as your users' browsers do, then times the text coming back.
This is what it found.
Checking the status…
Loading the latest checks.
- Uptime, 24 hours
- –
- Uptime, 7 days
- –
- Uptime, 90 days
- –
- Text back, 24 h
- – median
- – p95
- Connect, 24 h
- – median
- – p95
Every 5 minutes
Last 9030 days
- No failed checks
- 1–2 failed
- 3 or more failed
- No data
Show the days as a table
| Day (UTC) | Checks | Failed | Text median | Text p95 |
|---|
Latency, last 24 hours
- Text back: first audio sent → whole text received
- Connect: TLS and WebSocket, grant checked
- Failed check
No checks in the last 24 hours yet.
Show the checks as a table
| Time | Result | Connect | Text back |
|---|
What we measure
Every five minutes the control plane does what your product does, with an account of our own:
- Grant. It signs a short grant for 30 seconds of audio with that account's key, as your server would.
-
Connect. It opens a WebSocket to
wss://stream.transcribe.today/api/ws, over TLS and through the same front door as everyone. The node asks the control plane whether the grant is good before it lets the socket open. The time all of that takes is connect. -
Stream. It sends 5.5 seconds of Swedish speech, a sentence from a public-domain LibriVox
recording, as 16 kHz samples, as fast as the connection takes them, then
finish. - Text. It waits for the text and checks that it is the sentence's. Text back is the time from the first audio sent until the whole text has arrived.
A check that has not finished after 30 seconds counts as failed, at the step it was on. Uptime is the share of checks that went all the way through. The median is the typical time; p95 means 95 of 100 checks were at least that fast.
The checks run on the server that hosts the API, so the numbers show how the service itself is doing, not anyone's internet connection. The probe's account pays for its own seconds: no customer's credit is used.
The server was restarted recently and keeps its history in memory for now, so the longer periods only cover the time since then.
The same numbers as JSON: /api/status.