Cartesia for Enterprise Voice AI: Sonic Architecture & Integration

from cartesia import Cartesia
import os

def stream_cartesia_sonic_tts(text_prompt: str, voice_id: str = "a0e168ee-3a82-4414-a642-a3b5eb5d6515") -> None:
  """
  Streams sub-100ms text-to-speech audio using Cartesia Sonic model over WebSockets.
  """
  api_key = os.getenv("CARTESIA_API_KEY")
  client = Cartesia(api_key=api_key)

  print("Connecting to Cartesia Sonic WebSocket endpoint...")
  ws = client.tts.websocket()

  output_format = {
      "container": "raw",
      "encoding": "pcm_s16le",
      "sample_rate": 24000
  }

  # Execute streaming request
  for chunk in ws.send(
      model_id="sonic-english",
      transcript=text_prompt,
      voice_id=voice_id,
      output_format=output_format,
      stream=True
  ):
      # Process raw PCM audio bytes (send to audio output or WebRTC player)
      audio_bytes = chunk.audio
      print(f"Received audio chunk: {len(audio_bytes)} bytes")

  ws.close()

if __name__ == "__main__":
  prompt = "Hello! This is Cartesia Sonic streaming audio with sub-100 millisecond latency."
  stream_cartesia_sonic_tts(prompt)