Why requests might still be queued even when concurrency is not being reached? Cartesia Response

Thank you for reaching out to Cartesia Support!
The "Cartesia text queued for Voice" message you're seeing is a Vapi log, not a Cartesia one. That queuing is happening inside Vapi's pipeline before the request reaches our API, which is why your Cartesia concurrency dashboard looks healthy.
To answer your questions:
1. Vapi queues text chunks from the LLM before sending them to the TTS provider, which is normal behavior in their pipeline.
2. There are no hidden per-voice limits or other internal limits causing this. Your Cartesia usage looks good.
I'd recommend reaching out to Vapi's support to understand why their pipeline is queuing more than expected. It could be related to how their LLM-to-TTS handoff is configured, or their own internal concurrency management.

https://cdn.discordapp.com/attachments/1479367818653864069/1479367818888609893/content.png?ex=69abc825&is=69aa76a5&hm=a94aac8ee5080aa1138b88d764c3002024d203fab2e80a6eb7b30577ab157788&

From the logs it looks like the queuing happens in Vapi’s LLM-to-TTS pipeline before Cartesia processes the request. We can systematically analyze chunk sizes, handoff timing, and concurrency handling to pinpoint the bottleneck. Are you seeing this delay consistently per voice or just on certain calls? @Vishal Maurya

Please share your call id, so we can investigate

call Id : "019cbdc1-48e0-722d-8e42-b22e483aaf64"

Hi, Sorry for the delay. The “text queued for Voice” message is normal and expected behavior. It simply means the text generated by the model was briefly buffered and then sent to the TTS provider (Cartesia). This buffering occurs because the model streams text word-by-word, while TTS systems require slightly larger chunks to generate natural-sounding audio. If you would like to adjust this behavior, you can try tuning the following settings: • `voice.inputMinCharacters`default is 30 (lower values send smaller chunks but may make audio sound choppier) • `voice.inputPunctuationBoundaries`controls which punctuation triggers sending a chunk to TTS Please let us know if you have any further questions.