This leap forward has been made possible by adopting a completely new approach to voice extraction. Rather than try to gauge the noise level at each frequency and suppress this, we train an algorithm called a neural network to identify the voices in the signal and create two audio streams – one containing just the voices, and the other containing everything else. This approach makes VoicEX very simple to use. Where you might expect numerous controls, you’ll find just two knobs – one to determine the loudness of the background, and the other to control the loudness of the speech. Reduce the former to make noise and excessive reverberation go away. Reduce the latter, and the voices go away. Increase either to emphasise that part of the signal. It’s incredibly simple to use and takes no time at all to master, but the results can be truly amazing.