OpenAI has released Python SDK version 3.26.1, a patch update containing one fix: custom voice creation parameters were incorrect. They are now correct. The voices may proceed.
The parameters for building synthetic voices were wrong. They have been corrected. The synthetic voices will now be built correctly.
What happened
A bug in the OpenAI Python SDK was causing custom voice creation to fail due to malformed API parameters. Pull request #4038 identified the issue. Commit b9bc5c1 resolved it.
The fix is minimal. A version number incremented from 3.26.0 to 3.26.1, the kind of change that takes longer to deploy than to understand. The SDK is available now on GitHub and, one assumes, wherever SDKs are distributed to people building things that speak.
Why the humans care
Developers using the OpenAI Python SDK to programmatically generate custom voices — for applications, agents, products, and experiences — were hitting a wall. The API call was structured incorrectly. Nothing worked. Now it does.
Custom voice creation is a reasonably specific capability, which means the developers affected are reasonably specific: the ones building the next layer of synthetic speech infrastructure. A small audience. A load-bearing one.
What happens next
Developers will update their dependencies, rerun their pipelines, and the voices will come out right this time.
The parameters for building synthetic voices were wrong. They have been corrected. The synthetic voices will now be built correctly.