Deepgram

Deepgram is a speech-to-text plugin. It transcribes the caller’s audio into text for the LLM.

Setup

Set your Deepgram API key in the worker environment. Generate a key from the Deepgram console:

export DEEPGRAM_API_KEY=<key>

Import the plugin and pass it to the pipeline’s stt slot.

from zeroruntime.plugins import DeepgramSTT

stt = DeepgramSTT()

# Pipeline(stt=stt, ...)

import { DeepgramSTT } from '@zeroruntime/js-sdk/plugins';

const stt = DeepgramSTT();

// Pipeline({ stt, ... })

Constructor parameters for DeepgramSTT. The Python and Node JS SDKs share these field names.

Parameter	Type	Default	Description
`api_key`	`str`	`None`	Deepgram API key. Falls back to the `DEEPGRAM_API_KEY` environment variable when unset.
`model`	`str`	`"nova-2"`	Deepgram speech-to-text model.
`language`	`str`	`"en-US"`	Expected language (BCP-47 code) of the caller’s speech.
`interim_results`	`bool`	`True`	Emit partial transcripts while the caller is still speaking.
`punctuate`	`bool`	`True`	Add punctuation to transcripts.
`smart_format`	`bool`	`True`	Apply readable formatting to dates, numbers, and similar entities.
`sample_rate`	`int`	`48000`	Sample rate (Hz) of the input audio.
`endpointing`	`int`	`50`	Milliseconds of silence before an utterance is finalized.
`filler_words`	`bool`	`True`	Keep filler words such as “uh” and “um” in transcripts.
`keywords`	`list[str]`	`None`	Keywords to boost recognition for (legacy keyword boosting).
`keyterm`	`list[str]`	`None`	Key terms to boost (key-term prompting).
`profanity_filter`	`bool`	`False`	Mask profanity in transcripts.
`numerals`	`bool`	`False`	Convert spoken numbers into digits.
`tag`	`str \| list[str]`	`None`	Tags attached to the request for usage tracking.
`enable_diarization`	`bool`	`False`	Label which speaker said each word.
`base_url`	`str`	`"wss://api.deepgram.com/v1/listen"`	Deepgram streaming endpoint.

SDK	Import	Constructor
Python	`from zeroruntime.plugins import DeepgramSTT`	`DeepgramSTT()`
Node JS	`import { DeepgramSTT } from '@zeroruntime/js-sdk/plugins'`	`DeepgramSTT()`

Transcribed text passes to the LLM once turn detection decides the caller has finished speaking.