Potential Realtime TTS Ending Delay

[Kaister300]

Description

In the realtime TTS module, the convert_realtime module is not checking if the isFinal message has been received or not. This causes a couple hundred of milliseconds of delay between when the client receives the last message from the websocket and when the websocket is actually terminated. If this is intentional then that is fine but in order to get a more accurate reading on how long the entire response gets generated on the client side then I propose the following solution:

# src/elevenlabs/realtime_tts.py
        def convert_realtime(...):
                    # Line 138-ish
                    ...
                    if "audio" in data and data["audio"]:
                        yield base64.b64decode(data["audio"])  # type: ignore
                    if "isFinal" in data and data["isFinal"]:  # Checks if we receive final to exist generator function
                        break
                    ...

Code example

No response

Additional context

No response

[kraenhansen]

Thanks for reporting this, @Kaister300. You're right that the while True drain loop in convert_realtime doesn't check for the isFinal message from the server, so it waits for the server to close the connection instead of breaking early. Our own websocket documentation example actually does check for isFinal, so the SDK should be consistent with that. We'll look into adding the isFinal break condition you've proposed.

@cursor try to fix this