Synthesizers are crucial for designing sounds in today's music. However, to create the desired sound texture by tuning the right synthesizer parameters, one requires years of training and in-depth domain experience on sound design. Music producers might also search through preset banks, but it takes extensive time and effort to find the best preset that gives the desired texture.
Imagine a program that you can drop your desired audio sample, and it automatically generates the synthesizer preset that could recreate the sound. This task is commonly known as "parameter inference" of music synthesizers, which could be a useful tool for sound design. In this talk, we will discuss how deep learning techniques can be used towards solving this task. We will cover recent works that use deep learning to perform parameter inference on a variety of synthesizers (FM, wavetable, etc.), as well as the challenges that were faced in solving this task.
IF YOU ARE ATTENDING ONLINE, ALL TALK SESSIONS CAN BE ACCESSED FROM THE MAIN LOBBY:
https://conference.audio.dev
