Cartesia TTS

CartesiaTTS is a Patter TTSProvider backed by Cartesia’s /tts/bytes HTTP endpoint. It streams raw PCM_S16LE chunks that drop directly into Patter’s pipeline with no transcoding. Pure aiohttp transport, no vendor SDK required. The default model is sonic-3 (current GA snapshot, ~90 ms TTFB target).

This page covers Cartesia TTS. For Cartesia’s ink-whisper STT see the Cartesia STT page.

Install

pip install "getpatter[cartesia]"

Usage

Use the namespaced import (getpatter.tts.cartesia) or the flat alias (getpatter.CartesiaTTS). Both auto-resolve CARTESIA_API_KEY from the environment when api_key= is omitted.

# Namespaced import
from getpatter.tts import cartesia

tts = cartesia.TTS()                                      # reads CARTESIA_API_KEY
tts = cartesia.TTS(voice="f786b574-daa5-4673-aa0c-cbe3e8534c02")  # Katie

# Flat alias (equivalent)
from getpatter import CartesiaTTS

tts = CartesiaTTS(voice="f786b574-daa5-4673-aa0c-cbe3e8534c02")

// See typescript-sdk/providers/cartesia-tts for the TS twin.

In an agent:

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

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

agent = phone.agent(
    stt=DeepgramSTT(),                                    # DEEPGRAM_API_KEY from env
    tts=CartesiaTTS(voice="f786b574-daa5-4673-aa0c-cbe3e8534c02"),
    system_prompt="You are a helpful assistant.",
)

asyncio.run(phone.serve(agent))

Models and rates

Cartesia TTS bills per 1,000 characters synthesized. Per-model rates (defaults from getpatter.pricing):

Model	Rate / 1k chars	Notes
`sonic-3` (default)	$0.030	GA, ~90 ms TTFB. Drop-in compatible with sonic-2 voice IDs.
`sonic-2`	$0.030	Previous flagship.
`sonic-1` / `sonic-english`	$0.030	Legacy English.
`sonic-multilingual`	$0.030	Multilingual variant.

Override per-call via Patter(pricing={"cartesia_tts": {"models": {"sonic-3": {"price": ...}}}}).

Languages

language="en" by default. Sonic-3 supports 30+ languages — pass any Cartesia-supported BCP-47 code ("es", "fr", "de", "it", "pt", "ja", "zh", …). Voice cloning works across all of them — the same voice ID will speak whichever language you target.

Telephony optimization

Use the carrier-aware factories so audio reaches Patter’s telephony adapter at the carrier’s native sample rate, skipping a resample step:

from getpatter.tts import cartesia

# Twilio: PCM @ 8 kHz directly (skips the 16k→8k resample before mulaw transcode).
tts = cartesia.TTS.for_twilio()

# Telnyx: PCM @ 16 kHz (matches Telnyx's L16/16000 default — zero transcoding).
tts = cartesia.TTS.for_telnyx()

Options

Option	Default	Notes
`api_key`	`None`	Reads from `CARTESIA_API_KEY` when omitted.
`model`	`"sonic-3"`	Any Cartesia TTS model id.
`voice`	`"f786b574-..."` (Katie)	Cartesia voice id.
`language`	`"en"`	BCP-47 code.
`sample_rate`	`16000`	8000, 16000, 22050, 24000, 44100 Hz.
`speed`	`None`	`"fastest" ... "slowest"` or float in `[0.6, 2.0]`.
`emotion`	`None`	Cartesia emotion preset.
`volume`	`None`	Float in `[0.5, 2.0]` (sonic-3 only).
`base_url`	Cartesia API	Override for proxying.
`api_version`	`"2025-04-16"`	Cartesia API version pin.

Low-level usage

from getpatter.providers.cartesia_tts import CartesiaTTS

tts = CartesiaTTS(api_key="...", voice="f786b574-...", language="en")
async for chunk in tts.synthesize("Hello from the Patter pipeline."):
    ...
await tts.close()

​Cartesia TTS

​Install

​Usage

​Models and rates

​Languages

​Telephony optimization

​Options

​Low-level usage

Cartesia TTS

Install

Usage

Models and rates

Languages

Telephony optimization

Options

Low-level usage