How do I update the transcription_delay

Thanks for reaching out, Rithvik. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

Is it possible to male the transcription_delay longer?

Thanks — I’m responding to your question about whether you can make the

transcription_delay

longer. Short answer: The

transcription_delay

is a measured metric (the time between a user pausing and the finalized transcript appearing). You cannot directly set that metric to an arbitrary value. However, you can make transcripts appear later by changing the behavior that determines when transcription is finalized — for example, by increasing endpointing / turn-detection timeouts or adjusting STT provider settings that control when a transcript is emitted. To give exact steps I’ll need a bit more info: • Which STT provider are you using (AssemblyAI, Whisper, Google, etc.)? • Where are you configuring transcription behavior — in a LiveKit agent/turn-detector config, in your server code, or in the STT provider settings? • Do you want every transcript to be delayed by a fixed amount, or do you want to increase the pause/endpoint threshold so finalization waits longer? If you tell me the provider and where the config lives, I’ll provide concrete configuration options or sample settings to increase the effective transcription delay (for example, what endpointing/timeout parameter to change and where to set it). , with interactive elements

Deepgram

Thanks — you're asking how to make the

transcription_delay

longer when using Deepgram. Good news: you can increase the time before a transcript is finalized by adjusting Deepgram end-of-turn (EOT) settings in the LiveKit Deepgram STTv2 integration. What to change • eot_threshold (float): Higher values require greater confidence before finalizing a turn. Raising this increases delay. • eot_timeout_ms (number): Amount of silence (milliseconds) after which a turn is finalized regardless of confidence. Increase this to guarantee a longer maximum wait. • eager_eot_threshold (float): Threshold for an “eager” end-of-turn signal. Raising it will delay interim EOT signals. Example usages 1. Python (LiveKit Deepgram STTv2)

from livekit.plugins import deepgram

stt = deepgram.STTv2(
    model="flux-general-en",
    eot_threshold=0.8,       # increase for longer required confidence
    eot_timeout_ms=3000,     # wait up to 3s of silence before finalizing
    eager_eot_threshold=0.6,
)

1. TypeScript/JavaScript

const stt = new deepgram.STTv2({
    model: "flux-general-en",
    eotThreshold: 0.8,
    eotTimeoutMs: 3000,
    eagerEotThreshold: 0.6,
});

Guidance on choosing values • If you want transcripts to wait longer after a pause, raise

eot_timeout_ms

. • If you want stricter confidence before finalizing (which can increase delay), raise

eot_threshold

. • Use

eager_eot_threshold

to control interim/early EOT behavior. If you tell me: • The current values you’re using (if any), and • Whether you want a consistent fixed delay or to change behavior only for ambiguous/low-confidence speech, I can suggest concrete parameter values and a short test plan to verify the effect in your environment. Sources: Deepgram STT | LiveKit Documentation, with interactive elements