Deepgram STT Explorer

Input

Import / Export

Core

Model

Flux streams from /v2/listen. Batch mode and every v1-only parameter are unavailable, and the panel hides them.

Language

Base URL

Flux End-of-Turn

eot_threshold

0.5-0.9. Confidence needed to end a turn. Higher is more certain, slower.

eager_eot_threshold

0.3-0.9, unset by default. Setting it enables EagerEndOfTurn and TurnResumed. Must be ≤ eot_threshold.

eot_timeout_ms

500-60000. Silence that forces a turn to end regardless of confidence.

language_hint

flux-general-multi only. On flux-general-en it is rejected at the handshake.

Audio

Sample Rate

Channels

Encoding

Multichannel

Formatting

Smart Format

Punctuate

Numerals

Filler Words

Dictation

Profanity Filter

Word Confidence

Features

Diarize (v1, deprecated)

Detect Entities

Utterances

Paragraphs

Diarizer (diarize_model)

v2 is batch only. Deepgram rejects the WebSocket handshake for diarize_model=v2 on a stream; use latest there.

Sends diarize_model and omits diarize. The deprecated diarize checkbox always selects v1.

Replace

Redaction

Redact (select all that apply)

Prompting

Keyterms

Intelligence (batch only)

Topics

Intents

Sentiment

Streaming (live only)

Endpointing (ms)

Utterance End (ms)

Interim Results

VAD Events

No Delay

Advanced

Version

Alternatives

Callback URL

Tags (comma-separated)

MIP Opt-Out

Extra JSON params

This repo · deepgram · deepgram-devs · Docs · Community