Output stream issue

[WikiLucas00]

Hello! I tried this project but noticed some problems in the response stream. It seems the audio chunks are being played as soon as available, not waiting for the current one to finish playing.

Here's a video of the answer I received (using ./examples/streaming_cli.py) when I asked it to tell a story:

openai_realtime_client.mp4

[RazaProdigy]

I can confirm this. The audio chunks are not playing as expected.

[logan-markewich]

Interesting, if either @RazaProdigy or @WikiLucas00 can spot where the fix is, happy to merge a pr

I'll try to take a look soon if not

[mdagost]

Yeah--I'm seeing the same thing.

[mdagost]

In case it's helpful, this python code that I wrote does stream audio properly on my system: https://github.com/mdagost/openai-realtime-streamlit

[tanvithakur94]

This change helped me in fixing the above issue described - #5

[RazaProdigy]

@tanvithakur94 Still same issue after increasing queue max size ☹

[tanvithakur94]

@tanvithakur94 Still same issue after increasing queue max size ☹

@RazaProdigy If you use airpods or some input device with good noise cancellation, it should work. For better performance, you can mute the input volume after you are done talking.

[logan-markewich]

yea, in a noisy environment the interrupt mode is a little flakey. There is some threshold settings that you can play with, I don't think these are fully exposed in the API

I don't think the queue size has much to do with this

[ThreepE0]

This happens when the mic is muted after input. Similar experience as to when the model is picking itself up, but this issue is NOT related to that.