Bug: demo/gemini_audio_video/app.py gradio webpage shows Illegal header value b'Bearer '

[ahundt]

I was trying to finish up #88 and I started out by running the current functionality on the main branch (not the pull request's code):

demo/gemini_audio_video/app.py launches, but when I try to hit record on the video I get the following popup in the web ui (nothing prints to terminal):

Illegal header value b'Bearer '

It looks like it might relate to this line:

fastrtc/backend/fastrtc/credentials.py

Line 54 in a6a0937

"Authorization": f"Bearer {token}",

as that's the only instance of Bearer in fastrtc itself.

I already ruled out alternative causes like the gemini api key which is working. I think this is likely a fastrtc bug, but I'm not sure. Thanks for giving this a look.

[freddyaboulton]

Hi @ahundt ! Yea instead of using twilio for the TURN server, I am using cloudflare through a hugging face token. It's free if you have a hugging face account. You can just delete all mentions of get_hf_turn_credentials from the gemini script though if you're running locally.

I think that fixes your issue so I will close.

[ahundt]

Sorry @freddyaboulton could you elaborate how the switch to cloudflare resolves this particular issue with the bearer error?

This error appears in the main branch where the expected behavior is that the demo code runs successfully.

Maybe there is something implied like I have to set up a cloudflare account?

Given the bearer error encountered is very cryptic, could the code be updated to capture that sort of issue and provide guidance to users on what is wrong and what might be a reasonable next step to do?

Thanks!

[freddyaboulton]

Hi @ahundt ! Ah yes I see the issue. The function errors out if you don't have the HF_TOKEN or CLOUDFLARE_* environment variables set. I just fixed the demo so that that function doesn't run when developing locally: #332

[ahundt]

@freddyaboulton Cool, thanks for the explanation and the app fix!

Could there also be a fix in the upstream code itself so the code doesn't crash out too if the parameters are set as they were before? it seems like others might encounter the bearer bug in the future.

[freddyaboulton]

I think we should make it so that the function raises a more informative error if called without parameters and no env variables are set. I don't think we should fix by suppressing the error

[freddyaboulton]

Would you like to contribute that in a PR?