vercel/ai @ai-sdk/deepgram@2.0.71

on GitHub

Patch Changes

• 8dfbbed: fix (deepgram): reject the speech callback provider option instead of returning Deepgram's JSON acknowledgement as audio. With a callback URL, Deepgram processes the request asynchronously and answers with {"request_id": "..."} instead of audio, which generateSpeech previously returned as corrupt audio bytes. The provider now throws an InvalidArgumentError before sending the request, and warns when callbackMethod is passed without callback.

  If you relied on callback delivery, call Deepgram's /v1/speak REST API directly; generateSpeech always needs the audio in the response.