Deepgram STT

DeepgramSTT streams PCM (or mulaw / opus / flac) audio to Deepgram’s v1 /listen WebSocket endpoint. Pure websockets transport, no vendor SDK required. Handles KeepAlive pings, Finalize / CloseStream graceful shutdown, and surfaces both Results transcripts and SpeechStarted / UtteranceEnd VAD events.

Install

deepgram ships in the base install — no extra needed.

pip install getpatter

Usage

Use the namespaced import (getpatter.stt.deepgram) or the flat alias (getpatter.DeepgramSTT). Both auto-resolve DEEPGRAM_API_KEY from the environment when api_key= is omitted.

# Namespaced import
from getpatter.stt import deepgram

stt = deepgram.STT()                                      # reads DEEPGRAM_API_KEY
stt = deepgram.STT(api_key="dg_...", model="nova-3", language="en")

# Flat alias (equivalent)
from getpatter import DeepgramSTT

stt = DeepgramSTT()

// See typescript-sdk/providers/deepgram for the TS twin.

Plug it into an agent:

import asyncio
from getpatter import Patter, Twilio, DeepgramSTT, ElevenLabsTTS

phone = Patter(carrier=Twilio(), phone_number="+15550001234")

agent = phone.agent(
    stt=DeepgramSTT(),                                    # DEEPGRAM_API_KEY from env
    tts=ElevenLabsTTS(voice_id="rachel"),
    system_prompt="You are a helpful assistant.",
)

asyncio.run(phone.serve(agent))

Models and rates

Deepgram bills per minute of streamed audio. Per-model rates (defaults from getpatter.pricing):

Model	Rate / min
`nova-3` (default)	$0.0077
`nova-3-multilingual`	$0.0092
`nova-2`	$0.0058
`nova`	$0.0043
`whisper-large`	$0.0048
`whisper-medium`	$0.0048

Override per-call via Patter(pricing={"deepgram": {"models": {"nova-2": {"price": ...}}}}).

Languages

language="en" by default. Pass any Deepgram-supported BCP-47 code (e.g. "es", "fr", "de", "it"). For mixed-language calls use nova-3-multilingual.

Telephony optimization

For Twilio (mulaw 8 kHz) use the carrier-aware factory so audio reaches Deepgram natively without an extra resample step:

from getpatter.providers.deepgram_stt import DeepgramSTT as _LowLevelSTT

stt = _LowLevelSTT.for_twilio(api_key="dg_...")           # mulaw, 8 kHz

Options

Option	Default	Notes
`api_key`	`None`	Reads from `DEEPGRAM_API_KEY` when omitted.
`model`	`"nova-3"`	Any Deepgram model id.
`language`	`"en"`	BCP-47 code.
`encoding`	`"linear16"`	`"linear16"`, `"mulaw"`, `"alaw"`, `"opus"`, `"flac"`.
`sample_rate`	`16000`	8000, 16000, 24000, 44100, 48000 Hz.
`endpointing_ms`	`150`	Silence (ms) before Deepgram closes a turn.
`utterance_end_ms`	`1000`	Server-side endpoint detection (min 1000).
`smart_format`	`True`	Punctuation + numerals.
`interim_results`	`True`	Stream partial transcripts.
`vad_events`	`True`	Emit `SpeechStarted` / `UtteranceEnd`.

Low-level usage

from getpatter.providers.deepgram_stt import DeepgramSTT

stt = DeepgramSTT(api_key="dg_...", model="nova-3", language="en")
await stt.connect()
await stt.send_audio(pcm_chunk)                           # 16 kHz PCM s16le
async for t in stt.receive_transcripts():
    print(t.text, t.is_final, t.confidence)
await stt.close()

​Deepgram STT

​Install

​Usage

​Models and rates

​Languages

​Telephony optimization

​Options

​Low-level usage

Deepgram STT

Install

Usage

Models and rates

Languages

Telephony optimization

Options

Low-level usage