[BUG] StartConversationSessionAsync(model: "gpt-4o-mini-realtime-preview"), takes quite a few seconds to execute

[cjkarande]

Describe the bug

StartConversationSessionAsync(model: "gpt-4o-mini-realtime-preview") experiences significant initialization delay of 4-6 seconds, likely due to model loading occurring during conversation startup. This impacts the user experience with noticeable pauses at the beginning of each conversation session.

Current Behavior:
Conversation startup takes 4-6 seconds
Model appears to be loaded synchronously during StartConversationSessionAsync
Users experience significant delays before conversation begins

Expected Behavior
Faster conversation initialization
Model preloading capability to eliminate startup delays
Graceful session management

Proposed Solution

1. Model Preloading: Introduce a method to preload models at the RealTimeClient level before calling StartConversationSessionAsync
   // Suggested API
   var client = GetRealtimeClient("gpt-4o-mini-realtime-preview");
   // Model is preloaded here, making subsequent StartConversationSessionAsync calls faster

2. Graceful Session Termination: Add a dedicated StopConversationSessionAsync() method instead of relying solely on object disposal for session cleanup
   csharp
   // Suggested API
   await session.StopConversationSessionAsync();

Steps to reproduce

Call StartConversationSessionAsync with a specific model

realtimeSession = await realtimeClient.StartConversationSessionAsync(
model: "gpt-4o-mini-realtime-preview");//gpt-4o-realtime-preview

Code snippets

OS

Android/iOS (Xamarin Forms)

.NET version

.NET Standard 2.0

Library version

2.2.0

[jsquire]

Hi @cjkarande. Thank you for reaching out and we regret that you're experiencing difficulties. The path through the client for establishing a session is minimal, with only a small number of allocations via StartSessionAsync which then makes the call to establish the websocket connection. From the available context, there is not enough context available to suggest whether the delay you're experiencing is related to the application's host environment, network, or the service.

We appreciate your feedback and suggestions for the client API. Unfortunately, neither is something that we plan to move forward with at this time. The real time API surface is still in beta for the service and, likewise, experimental on the client. We will be sure to take your suggestions into account as we iterate on designs going forward.

With respect to your suggestion to preload models, the client has no insight nor influence over how models are managed by the service, and the service offers no API that would enable the client to support your suggestion. If the service should choose to expose an operation to do so, the client will support it. Your best path forward to suggest a servicve feature, if you would like to do so, is reaching out to the OpenAI Support Team. You can either email support@openai.com or start a new chat session via the Help Center. The OpenAI support team is available 24/7 and is equipped to assist with the full range of requests.

At present, the intent is that a session cannot be reused which is why we believe that Dispose is the right pattern. Introducing a StartConversationSessionAsync would potentially be confusing, as it gives the impression that you can call StartConversationSessionAsync and begin using it again. While that may be a pattern that we allow in the future, it is currently not one that we current support or plan to introduce in the short term.

[jsquire]

As there has been no follow-up discussion or questions in the last two weeks, closing this out.