Precisely timestamping when audio leaves speaker

One of the requirements I have is to precisely timestamp when audio leaves the speakers. I've tried using OnAudioFilterRead, the Update loop and a separate thread to measure this, but am not able to get any reliable measure of when audio leaves the speaker. (Standard deviation around 11-15 ms with the various approaches I've tried) I need to get the standard deviation sub 5 ms.

I am using AudioSource.PlayScheduled with AudioSettings.dspTime to schedule the audio, and I've reduced the buffer size for best performance / lowest latency.

Is it possible to precisely timestamp when audio is audible for the user, within 5 ms of error using Unity audio? Any hints on what needs to be done?

I am using Unity 2018.1.6f1