TTS: Audio generation stops when sending second text with `continue_=True` on same context

[mihbt]

I'm experiencing an issue where sending a second text chunk to the same TTS context causes the ongoing audio generation to immediately stop, even when using the continue_ parameter correctly.

Expected Behavior

When sending multiple text chunks to the same context with continue_=True, the audio should continue generating sequentially for all chunks.

Actual Behavior

The audio generation for the first text stops immediately when the second context.send() is called.

Environment

Cartesia SDK Version: 2.0.17
Python Version: 3.13.2

Code Example

from cartesia import AsyncCartesia, OutputFormat_RawParams, TtsRequestIdSpecifierParams

# Setup
client = AsyncCartesia(api_key="...")
ws = await client.tts.websocket()
context = ws.context("my-context-id")

# First text - this works fine
await context.send(
    model_id="sonic-3",
    transcript="First text to convert to speech.",
    voice=TtsRequestIdSpecifierParams(mode="id", id="ac197a78-cec7-4c50-93e5-93bdc1910b11"),
    stream=True,
    output_format=OutputFormat_RawParams(
        container="raw",
        encoding="pcm_s16le",
        sample_rate=22050,
    ),
    continue_=False  # First message
)

# Second text - this causes the audio to stop generating
await context.send(
    model_id="sonic-3",
    transcript="Second text to convert to speech.",
    voice=TtsRequestIdSpecifierParams(mode="id", id="ac197a78-cec7-4c50-93e5-93bdc1910b11"),
    stream=True,
    output_format=OutputFormat_RawParams(
        container="raw",
        encoding="pcm_s16le",
        sample_rate=22050,
    ),
    continue_=True  # Continuation of previous
)

await context.no_more_inputs()

# Receiving audio
async for output in context.receive():
    if output.audio:
        # Audio for second text never arrives
        handle_audio(output.audio)

Observations

• The first text generates audio successfully
• The audio generation stops exactly when context.send() is called for the second text. (ex: if I add a 1s delay - it will stream the audio output for 1s and after that stop)
• Waiting for the first audio to be completely generated before sending the second text doesn't help either, the second text is not generated at all.
• The continue_ flag is set correctly (False for first, True for subsequent)
• I'm receiving audio in a concurrent task that starts before sending any text

Questions

1. Is it expected to call context.send() multiple times on the same context?
2. Does continue_=True require a specific timing or pattern between sends?
3. Should I be using separate contexts for each text chunk instead? If yes, how do I keep the voice consistency/prosody between them?
4. Is there a way to queue multiple text chunks for sequential processing? (to keep voice consistency/prosody)

My use Case

I'm streaming LLM-generated responses to TTS and need to send chunks as they arrive for minimal latency. I want to maintain voice consistency across chunks, which is why I'm trying to use the same context with continue_=True.

Any guidance would be greatly appreciated! 🙏

[cheesecake100201]

Need help for the same @noahlt @bpanahij

[mihbt]

@cheesecake100201 Usingcontinue_=True for all the text chunks, including the first one fixed it for me. @chongzluong reply #58 (comment) brought me on the right path, thanks for it.

@cheesecake100201 if it works for you as well I can close this thread.