AssemblyAI documents a compatible speech API path ahead of OpenAI’s January TTS retirement

AssemblyAI documented a compatible API path ahead of an announced speech model retirement.

**Dek:** AssemblyAI’s November 3 guide shows how to keep an existing SDK call while changing the service behind it. The proposed migration preserves the request shape, with voice selection and streaming behavior still requiring review.

AssemblyAI published a migration guide on November 3 for developers using OpenAI’s dedicated speech endpoint. OpenAI has scheduled the retirement of tts-1, tts-1-hd, gpt-4o-mini-tts and listed snapshots for January 6, 2027. Its API reference directs new integrations to gpt-realtime-2.1-mini through the Realtime API. AssemblyAI offers a separate path through an endpoint that accepts the existing speech request shape, with a different base URL, API key, model and voice. [AssemblyAI migration guide](https://www.assemblyai.com/blog/migrate-openai-tts-to-assemblyai) · [OpenAI deprecations](https://developers.openai.com/api/docs/deprecations)

The endpoint is only one part of the contract

A team can keep using the SDK’s speech call while sending requests to AssemblyAI. The migration guide provides code examples and maps voice choices. An application also needs to confirm what it sends today and what its playback stack expects to receive. A voice selected by name may have no exact equivalent. Audio formats, chunk timing and error behavior can affect a voice application even if the same SDK method still returns bytes. [AssemblyAI migration guide](https://www.assemblyai.com/blog/migrate-openai-tts-to-assemblyai)

The provider change also changes where text is sent and which credentials authorize requests. That matters when prompts include customer information or messages written for a private service. An operator should inventory which service accounts call the existing endpoint and verify where the new credential is stored and rotated. Credentials for the new service belong with that service’s configuration. A code change that only alters the base URL can leave unresolved questions about data handling and operational ownership.

Compare the two integration paths on a representative flow

For a prerecorded narration job, the acceptance test can compare output format, duration, intelligibility and the application's handling of an invalid request. For an interactive voice agent, time to first audio and interruption behavior can matter more. The team can use the same short set of prompts when comparing a compatible endpoint with a move to Realtime, and document which parts of its workflow need to change in each case. [AssemblyAI migration guide](https://www.assemblyai.com/blog/migrate-openai-tts-to-assemblyai) · [OpenAI speech API reference](https://developers.openai.com/api/reference/resources/audio/subresources/speech/methods/create)

The January date gives operators a deadline to locate old requests and plan the transition. AssemblyAI’s guide shows an alternative with a familiar call shape. A successful migration still depends on the provider matching the application’s audio, latency and service requirements.