< Back to IRCAM Forum

Normalizing input for mfcc analysis for granular audio mosaicing?

Hello there! I am running through lots of questions regarding what is best for audio mosaicing. I know it all depends on each particular situation so I will describe the setting as exactly as I am able to, hoping you could help me.

I am designing a Max patch for perfoming live audio granular mosaicing on a woodwind quintet using MFCC. Each instrument has its own microphone and is routed to an individual channel through the audio interface. Before starting playing the music, we always perfom a brief sample recording of each instrument to use as target for the KNN selection based on live input.

My question is whether I should go through normalization both while recording the target and then in live input before the live MFCC analysis. May this affect mosaicing in any way?

I try to keep the gain of the mics at the audio interface level quite low to avoid capturing noise from the other instruments and ambient noise that triggers mosaicing.

I hope I have been clear enough. Any help will be much appreciated!