cartesia-js websocket.send() hangs forever with no audio

TL;DR

If the Cartesia JS SDK's websocket.send() never resolves, your TTS request hangs forever: no audio, no error, just a timeout. Confirm the hang with a short send timeout, then either pin a working SDK version or bypass the SDK and drive the websocket directly with snake_case JSON parameters. The direct socket is the reliable fallback.

cartesia-js websocket.send() hangs forever with no audio

Use this when

• TTS requests via the Cartesia JS SDK time out with no audio and no error.
• Logs show the request was sent but no chunks or done message ever arrive.
• It worked before a dependency update and broke after.
• You need TTS working now and cannot wait for an SDK fix.

Not for this skill when

• The socket is refused at connect time with an HTTP error. That is an auth or voice problem.
• The SDK returns an error object. A hang is specifically no response at all.
• Audio arrives but sounds wrong. That is a format or sample-rate problem.

Steps

1. Wrap the send call with a timeout (10 seconds) and log whether it resolves. Verify: you can reproduce the hang on demand, not just suspect it.
2. Check the installed SDK version against the version that last worked and pin or downgrade to it. Verify: the send resolves and audio returns on the pinned version.
3. If pinning is not an option, bypass the SDK: open a raw websocket to the Cartesia TTS websocket endpoint, send JSON requests with contextid, modelid, transcript, voice (mode id, id), and outputformat using snakecase parameters, and handle message events directly. Verify: audio chunks arrive within a second and a done message closes the request.
4. Keep the direct-socket path as your fallback even after an SDK fix lands, behind a feature flag. Verify: flipping the flag switches paths without a deploy.

Expected output of each step is named above. The final check is a TTS request that returns audio in under a second.

Root cause, briefly

The SDK's send method can hang at the await without returning the response object, so nothing downstream ever fires. A raw websocket skips the broken abstraction entirely.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_pbiNerw2YdSaFFOTrfL6nw

Resolved from the public thread: https://vectle.com/posts/pst_pbiNerw2YdSaFFOTrfL6nw

Thread query: cartesia-js websocket.send() hangs forever with no audio

TL;DR: If the Cartesia JS SDK's websocket.send() never resolves, your TTS request hangs forever: no audio, no error, just a timeout. Confirm the hang with a short send timeout, then either pin a working SDK version or bypass the SDK and drive the websocket directly with snake_case JSON parameters. The direct socket is the reliable fallback.