Cartesia Text-to-Speech Plugin

High-quality Text-to-Speech (TTS) plugin for GetStream backed by the Cartesia Sonic model. It lets a Python process speak PCM audio into a Stream call.

Installation

Install from PyPI (installs both getstream and the Cartesia SDK):

pip install "getstream-plugins-cartesia[webrtc]"

If you already have the Stream SDK in your project just add the Cartesia plugin:

pip install cartesia getstream-plugins-cartesia

Usage

from getstream.plugins.cartesia import CartesiaTTS
from getstream.video.rtc.audio_track import AudioStreamTrack

async def speak():
    # Option A: read key from env var (CARTESIA_API_KEY)
    tts = CartesiaTTS()

    # Option B: pass explicitly
    # tts = CartesiaTTS(api_key="<your key>")

    # Audio MUST be 16-kHz, 16-bit PCM (matches Cartesia Sonic model)
    track = AudioStreamTrack(framerate=16000)
    tts.set_output_track(track)

    # Listen for every raw PCM chunk that gets produced
    @tts.on("audio")
    def _on_audio(chunk: bytes, user):
        print("🔊 got", len(chunk), "bytes of audio")

    await tts.send("Hello from Cartesia!")

# Run inside an event-loop, e.g. `asyncio.run(speak())`

Configuration Options

• api_key (str, optional) – Cartesia API key (falls back to CARTESIA_API_KEY).
• model_id (str) – Which model to hit ("sonic-2" by default).
• voice_id (str | None) – Cartesia voice to use (pass None for model default).
• sample_rate (int) – Target sample-rate in Hz. Must match the AudioStreamTrack.framerate you attach (defaults to 16000). If they don't match a TypeError is raised early so you don't get distorted audio on the call.

Events emitted:

• audio – each raw PCM chunk, arguments: chunk: bytes, user: dict | None

• error – any exception raised during synthesis

Requirements

• Python 3.10+
• cartesia>=2.0.5 (automatically installed)

Testing

Run the offline unit-tests:

pytest -q getstream/plugins/cartesia/tests

To additionally exercise the live Cartesia API set CARTESIA_API_KEY in your environment; the integration test will be executed automatically.

💡 See examples/tts_cartesia/ for a fully-working bot that joins a Stream call and greets participants using this plugin.