Migration guide from Beta

If you are using the deprecated OpenAI realtime client in your application, this guide will help you to migrate to the newest version of the realtime client.

Follow the steps below to migrate to the GA version:

1. Remove the Beta namespace for the GA RealtimeAPI Client. For example, OpenAI.Beta.functionName becomes OpenAI.functionName, the same for events, etc.
2. RealtimeAPIClientParameters now have the type parameter that lets developers choose between two API modes (realtime/transcription).
3. RealtimeAPIClient methods related to the API communication (sessionUpdate, responseCreate, conversationItemCreate, etc.) expect parameters to be a full object as it is described in the OpenAI Realtime API reference, except the "type" field on the top level, which is determined by the method name itself.

For example, in the Beta Client sessionUpdate was called in the following manner:

1
realtimeAPIClient.sessionUpdate({
2
    "instructions": "You are a helpful assistant.",
3
    "voice": "sage",
4
    "input_audio_transcription": {
5
        "model": "whisper-1"
6
    }
7
});

In the GA Client sessionUpdate is called as follows:

1
realtimeAPIClient.sessionUpdate({
2
    "session": {
3
        "type": "realtime",
4
        "instructions": "You are a helpful assistant.",
5
        "audio": {
6
            "input": {
7
                "transcription": {
8
                    "model": "whisper-1",
9
                    "language": "en"
10
                }
11
            }
12
        }
13
    }
14
});

4. Several of the OpenAI messages and parameters have been renamed in GA, so please follow the OpenAI Realtime API Reference when creating parameters objects for the GA Client methods. More info is also available at https://platform.openai.com/docs/guides/realtime#beta-to-ga-migration.
5. Interruption support requires additional VoxEngine code to clear the media buffer when input_audio_buffer.speech_started event is received from OpenAI’s server VAD:

1
realtimeAPIClient.addEventListener(OpenAI.RealtimeAPIEvents.InputAudioBufferSpeechStarted, (event) => {
2
    if (realtimeAPIClient) realtimeAPIClient.clearMediaBuffer();
3
});