oviano
Yes. generate.js is place where the Avisynth script is created. Chainik knows more about internal Avisynth buffer than me. He knows why 'do not try 3rd option' 
In any way SVP needs video buffer to process frames and to produce intermediate ones. Video delay is a fact.
So you need to find a way to add the same delay to audio.
I don't know why it is not important when play file from disk and important when capture video and process frames on-the-fly.
Maybe answer is hidden here.
Audio track not processed by SVP. It is even not used in AVS script. Audio data is bypassing from ffdShow and Avisynth. So you need separate solution for audio.
