Occasional empty audio streams while sending subsequent requests

[austere panther]

Hi Cartesia folks,

We're building an interactive audio theater platform (Rolely) and we'd like to have multiple characters speak their dialog. We're doing this by making subsequent requests to the API using the SSE endpoint with timestamps and reading the output stream. This works nearly all of the time; however, around 10-20% of the time, some of the last few audio streams are completely empty (no logged bytes). I've validated that they're indeed empty from Cartesia and we're not hitting concurrency limits.

A minimal example of this is as follows (note this works most of the time, but fails roughly once out of 10 times with maybe the first 3 streams returning bytes and the last 2 being empty): https://gist.github.com/umairnadeem/3581632add44b83f3d72da9df8a68162

Has anyone else faced this same problem? Thank you for your help!

[open pumice]

Hi! I ran the script you shared ~10 times and was unable to reproduce empty output streams. Do you have any context ids of the failing requests? Are you still seeing this issue?

[austere panther]

thank you for replying! did you replace the embedding ids with actual embeddings? we're not passing any context ids

[austere panther]

Note that we are still indeed seeing this issue but it's intermittent and random. Would it be easier if we schedule a call?

[open pumice]

Yes, I had replaced the embedding ids with embeddings. I would love to hop on a call, when are good times for you?

[austere panther]

I can do:

• 10a-12p today
• 1-3p Tuesday
• 10a-12p Wed

What time works best for you?

[open pumice]

is this in pacific time?

[austere panther]

Yes

[open pumice]

I can do anytime between 2-3 PM Tuesday, feel free to send me a calendar invite [email protected]

[austere panther]

Cool sent you an invite