Audio Enhancement on Room Recording - is it Raw or Processed?

[Pawel_Lach (Pawel Lach (ai-coustics))]

Hey! I was just wondering when I am recording a room audio on LK with some audio enhancement plugins, will it be included in the recording or does the recording takes the raw input?

In other words - if a participant publishes their mic with an AudioProcessorapplied (e.g. noise cancellation), does room composite egress record the post processed track, or the raw input?

Quick test on my side with a noise cancellation plugin (ai-coustics) and ventilation fan as background noise. Three spectrograms attached:

• spec_raw: mic tapped before the processor

  

• spec_enhanced: mic tapped after the processor

  

• spec_egress: room composite egress

  

Noise floor in the egress gaps looks closer to raw than enhanced, so it seems egress captures the pre processed audio. Can someone confirm, and is there a way to configure egress to record the post processed track instead?

Thanks!

[darryncampbell (Darryn)]

Are you applying audio enhancement in the Agent, i.e. Noise & echo cancellation | LiveKit Documentation ?

If so, that noise cancellation will not be applied to the egress track.

[_E_r_r_o_n_e_o_u_s_D_e_s_i_g_n_s]

The way it works, Mic → (optional AudioProcessor) → encoded → sent to SFU

Server side (SFU / egress) → Received encoded track → mixed/composited → recorded (egress)

The key detail is Egress only has access to the published track, not your local pre/post processing tap.

Egress records the processed version only if, the processed audio is what’s actually being published as the track, meaning, you replace the original mic track with a processed one, or an easy way to see it is your processor modifies the outgoing MediaStream before publishing it.

Hope it helps.

[Pawel_Lach (Pawel Lach (ai-coustics))]

Great! Thank you for the replies and let’s keep it here just in case.