Between the moment someone speaks and the moment a listener reads or hears it, each step adds a little time. Some of it is the network; some of it is InterScribe waiting to understand the sentence before translating it.
The listener uses
Mic
Up
Words
Translate
Voice
Down
- CaptureThe agent picks up the sound and sends it in small pieces.
- Network upFrom your venue to InterScribe. Longer on Wi-Fi, satellite or a busy network.
- Recognising speechTurning sound into words, then confirming them as the sentence continues.
- TranslatingWaits for enough of the sentence to keep the meaning right.
- Speaking (AI voice)The voice starts once there are enough words to say naturally.
- Network downFrom InterScribe to the listener’s phone, over venue Wi-Fi or mobile data.
The longest path: voices wait for whole phrases so they sound natural. Your own numbers are in each session’s reliability report.
What you can shorten
- The network: cable the agent computer (why)
- Listener devices: wired headphones instead of Bluetooth
- The source: use the agent rather than RTMP, SRT or WHIP for in-room events
- Understanding the sentence before translating it, so the meaning is right
- A voice waits for enough words to speak naturally
More on each step
- The network: Fast is not the same as steady explains how to measure your connection.
- Listener devices: Bluetooth headphones add their own delay. Wired headphones or the phone's speaker avoid it. See Listening on a phone.
- The source: a stream you already produce arrives a few seconds after the room, which suits a broadcast. See Get good audio into InterScribe.
- Speech recognition shows words as they are heard, then confirms them as the sentence continues.
See the delay in the Console
Next to each channel, the Console shows its lag: how long that channel takes to follow the floor's newest caption. A few seconds is normal. A channel that reaches 3 s reads behind, and one that reaches 15 s reads down. See Captions arrive late.