Pipecat plugin

pipecat-slng adds speech to text (STT) and text to speech (TTS) services for Pipecat. It routes your pipeline through the SLNG gateway, so any streaming model on the platform works behind one API key. Swap the model string to switch provider; no other code changes. This page is the plugin reference. For a walkthrough that sets up your models in the dashboard and applies the change, see the Pipecat integration guide.

Tested with Pipecat v1.8.0. world_part requires pipecat-slng 0.6.0 or later. Upgrading from an earlier release? See Move from base_url.

Prerequisites

Python 3.11+ and pipecat-ai>=1.8.0
A Pipecat project
An SLNG API key. See Create your API key.

Install

uv add pipecat-slng
# or: pip install pipecat-slng

Quickstart

Each service reads your key via api_key and needs a world_part, the region that handles its requests. Create an STT and a TTS service and add them to your pipeline:

import os

from pipecat_slng import SlngSTTService, SlngTTSService

stt = SlngSTTService(
    api_key=os.getenv("SLNG_API_KEY"),
    world_part="us-east",
    model="slng/deepgram/nova:3-en",
)
tts = SlngTTSService(
    api_key=os.getenv("SLNG_API_KEY"),
    world_part="us-east",
    model="slng/deepgram/aura:2-en",
    voice="aura-2-thalia-en",
)

Both services stream over WebSocket, with low latency and mid-utterance interruption. Common runtime knobs are top-level keyword arguments (language, speed, enable_partials). For richer overrides, pass SlngSTTSettings(...) or SlngTTSSettings(...) to settings=.

Full voice agent example

A cascade pipeline (STT → LLM → TTS) using SLNG for speech and OpenAI for the language model. The full version, including the Daily transport, lives in examples/bot.py.

bot.py

import os

from pipecat.audio.vad.silero import SileroVADAnalyzer
from pipecat.frames.frames import LLMRunFrame
from pipecat.pipeline.pipeline import Pipeline
from pipecat.pipeline.runner import PipelineRunner
from pipecat.pipeline.task import PipelineParams, PipelineTask
from pipecat.processors.aggregators.llm_context import LLMContext
from pipecat.processors.aggregators.llm_response_universal import (
    LLMContextAggregatorPair,
    LLMUserAggregatorParams,
)
from pipecat.services.openai.responses.llm import OpenAIResponsesLLMService
from pipecat.transcriptions.language import Language
from pipecat.transports.base_transport import BaseTransport

from pipecat_slng import SlngSTTService, SlngTTSService


async def run_bot(transport: BaseTransport):
    key = os.environ["SLNG_API_KEY"]
    world_part = os.getenv("SLNG_WORLD_PART", "us-east")

    stt = SlngSTTService(
        api_key=key, world_part=world_part, model="slng/deepgram/nova:3-en",
        language=Language.EN, enable_partials=True,
    )
    tts = SlngTTSService(
        api_key=key, world_part=world_part, model="slng/deepgram/aura:2-en",
        voice="aura-2-arcas-en", language=Language.EN,
    )
    llm = OpenAIResponsesLLMService(api_key=os.getenv("OPENAI_API_KEY"))

    context = LLMContext()
    user_agg, assistant_agg = LLMContextAggregatorPair(
        context,
        user_params=LLMUserAggregatorParams(vad_analyzer=SileroVADAnalyzer()),
    )

    pipeline = Pipeline([
        transport.input(), stt, user_agg, llm, tts, transport.output(), assistant_agg,
    ])
    task = PipelineTask(pipeline, params=PipelineParams(enable_metrics=True))

    @task.rtvi.event_handler("on_client_ready")
    async def on_client_ready(rtvi):
        context.add_message({"role": "user", "content": "Please introduce yourself."})
        await task.queue_frames([LLMRunFrame()])

    await PipelineRunner(handle_sigint=False).run(task)

Model identifiers

Models follow provider/model:variant, with a -lang suffix where the model exposes per-language variants. Prefix with slng/ for an SLNG-hosted instance; a bare identifier is the external passthrough route, required for BYOK.

model="slng/deepgram/nova:3-en"      # SLNG-hosted (STT)
model="deepgram/nova:3"              # external passthrough (STT)

The full list is the model catalog. Not every model accepts every option (for example speed on TTS); check the Unified TTS reference before tuning.

STT reference

SlngSTTService streams speech to text over WebSocket, connecting to wss://{region}.api.slng.ai/v1/bridges/unmute/stt/{model}. Partial transcripts below 0.5 confidence are dropped before reaching your pipeline. Final transcripts are always kept. Language is imported from pipecat.transcriptions.language.

Parameter	Default	Description
`api_key`	required	SLNG API key.
`world_part`	required	Region code, as a string or `WorldPart` (see Region routing).
`model`	`"slng/deepgram/nova:3-en"`	Model identifier.
`encoding`	`"linear16"`	`"linear16"`, `"mp3"`, or `"opus"`.
`sample_rate`	pipeline rate	Audio sample rate in Hz.
`language`	model default	Recognition language. When you omit it, the gateway uses the model’s default language.
`enable_partials`	`True`	Stream interim transcripts. Only models that support partials use it.
`provider_key`	`None`	Your own provider key (BYOK, external routes only).
`settings`	`None`	`SlngSTTSettings` for runtime updates.

TTS reference

SlngTTSService streams text to speech over WebSocket, connecting to wss://{region}.api.slng.ai/v1/bridges/unmute/tts/{model}. Changing voice, speed, or language mid-session reconnects the WebSocket to re-run the init handshake. Pick a voice from the text to speech catalog.

Parameter	Default	Description
`api_key`	required	SLNG API key.
`model`	`"slng/deepgram/aura:2-en"`	Model identifier.
`world_part`	required	Region code, as a string or `WorldPart` (see Region routing).
`voice`	`None`	Voice identifier (server default if unset).
`encoding`	`"linear16"`	`"linear16"`, `"mp3"`, `"opus"`, `"mulaw"`, or `"alaw"`.
`sample_rate`	pipeline rate	Audio sample rate in Hz.
`language`	`Language.EN`	Synthesis language.
`speed`	`None`	Speech speed multiplier (server default if unset).
`provider_key`	`None`	Your own provider key (BYOK, external routes only).
`pronunciation`	`None`	A pronunciation dictionary for the session, for example `{"mode": "rewrite", "name": "support-pronunciations"}`.
`warm_standby_enabled`	`False`	Prepare one spare connection for the next utterance. It opens an extra provider session, which can use connection quota or cost money.
`settings`	`None`	`SlngTTSSettings` for runtime updates.

Region routing

world_part picks the region that handles a service’s requests. The service connects to {region}.api.slng.ai, and requests are not routed between regions. Each service takes its own world_part, so STT and TTS can run in different regions. Pass the region code as a string or as a WorldPart value. This example runs Soniox STT and Fish TTS in Germany:

import os

from pipecat.transcriptions.language import Language
from pipecat_slng import SlngSTTService, SlngTTSService, WorldPart

stt = SlngSTTService(
    api_key=os.getenv("SLNG_API_KEY"),
    world_part=WorldPart.EU_WEST,  # same as "eu-west"
    model="soniox/speech-ai:rt-v5",
    language=Language.EN,
)
tts = SlngTTSService(
    api_key=os.getenv("SLNG_API_KEY"),
    world_part="eu-west",
    model="slng/fish/tts:s2.1-pro",
    voice="1258154bb41540c7b9dc0d10cd6704bf",
    language=Language.EN,
)

Models differ by region. See Regions and base URLs for every region code and models by region for the speech models each region serves.

Move from `base_url`

Releases before 0.6.0 picked the region with base_url. Remove base_url and pass world_part instead. base_url still works if its host matches world_part, but it logs a DeprecationWarning:

# Works, with a warning. Remove base_url.
SlngSTTService(api_key=key, world_part="in", base_url="in.api.slng.ai")

# ValueError: the global api.slng.ai gateway is no longer supported.
SlngSTTService(api_key=key, world_part="in", base_url="api.slng.ai")

# ValueError: the host does not match world_part.
SlngSTTService(api_key=key, world_part="us-west", base_url="in.api.slng.ai")

0.6.0 also removes these:

region_override and world_part_override. Use world_part.
enable_vad on SlngSTTService. The service no longer sends it, and it never changed transcription. Delete the argument.
SlngHttpTTSService. Use the streaming SlngTTSService.

STT no longer sends language="en" by default. Pass language=Language.EN if you relied on that default on a multilingual model.

Bring your own key (BYOK)

Pass provider_key to bill an upstream provider directly on your own contract; both services forward it as the X-Slng-Provider-Key header, and the SLNG cache still applies on top. See Bring your own key.

stt = SlngSTTService(
    api_key=os.getenv("SLNG_API_KEY"),
    world_part="us-east",
    model="deepgram/nova:3",            # external route, no slng/ prefix
    provider_key=os.getenv("SLNG_PROVIDER_KEY"),
)

BYOK works on external routes only (no slng/ prefix). SLNG-hosted routes reject the header with an HTTP 400. Since 0.4.0, WebSocket connect-rejection errors include the server’s response body, so a misrouted request reports the reason.

Next steps

Pipecat integration guide for the migration walkthrough.
The model catalog and Unified API guide.
Using LiveKit instead? See the LiveKit plugin.

Unmute

SDKs

Skills

AI tools

Plugins

Prerequisites

Install

Quickstart

Model identifiers

STT reference

TTS reference

Region routing

Move from `base_url`

Bring your own key (BYOK)

Next steps

​Prerequisites

​Install

​Quickstart

​Model identifiers

​STT reference

​TTS reference

​Region routing

​Move from base_url

​Bring your own key (BYOK)

​Next steps

Prerequisites

Install

Quickstart

Model identifiers

STT reference

TTS reference

Region routing

Move from `base_url`

Bring your own key (BYOK)

Next steps