Issue link: http://digital.copcomm.com/i/1351516
SPECTRAL EDITING Next came the spectral editor, which was born out of a CEDAR project investigating better ways to de-click audio and remove other unwanted short duration events. Reid explains, "We had found that the spectral power of some clicks and related disturbances was constrained in the frequency domain and that we could perform a more accurate repair if we only processed the damaged frequencies rather than the whole signal in the time domain. This led to the definition of a click in both frequency and time, and it was then a short jump to removing the unwanted sounds of dropped microphones, slammed doors, coughs during the otherwise perfect take, and everything else. We patented this technology, and every spectral editor that you can buy today is based upon technology that is licensed (or should be licensed) from CEDAR." CEDAR released Retouch in February of 2002, and it was the first time the world had witnessed spectral editing. With a spectrogram, audio could now be presented in three dimensions: time, amplitude, and frequency. iZotope released its first edition of RX in June of 2010. Incorporating some spectral editing technology licensed from CEDAR, this was offered at a competitive price point and, for many mixers at the time, was their first introduction to the world of complex noise reduction and spectral repair. iZotope offered quick and easy-to-use tools to complete tasks such as de-clicking, de-crackling, and broadband noise reduction. As each new edition was released, tools such as De-Rustle, De-Wind, and De-Plosive made their way into the system. With the release of RX6 in 2018, iZotope introduced Dialogue Isolate. Dialogue Isolate uses machine learning to recognize and separate spoken dialogue from non-stationary background noises. When interviewed in the Spring 2020 edition of CAS Quarterly, Mike Rozett, iZotope's principal product manager, had this to say about the utilization of artificial intelligence (A.I.), "I would say that machine learning is getting to be a bigger and bigger part of what we're doing… It has allowed us to solve problems in a way that you couldn't do before. Artificial intelligence has helped a lot of industries innovate, and it's no exception in audio. Machine learning allows you to do some incredible things like separate a source. There were some EQ tricks you could do to separate different bands and side processing—and they were pretty effective at the time. But with machine learning, you can take a file and break things into parts and stems. That's still really early in its implementation and I think, going down the road with it, we're going to do some things that even a few years ago seemed impossible." [Of note, I used the Dialogue Isolate feature extensively on a recent series this summer where much of the clip-based source footage contained existing music beds under the dialogue that could not be licensed. With Dialogue Isolate and some creative rebuilding, dozens of clips were able to be used. –Ed.] Today, there are many entries to consider alongside CEDAR and iZotope, such as Accusonus ERA Noise Remover, Brusfri by Klevgrand, Audionamix IDC (Instant Dialogue Cleaner), Acon Digital's Extract: Dialogue, and Zynaptic's "Un" series. As new tools come to market, we have more opportunities to do things we once deemed impossible. However, the noise-reduction story is certainly far from over. It does appear that machine learning will likely continue to play a huge role in completing the complex mathematical equations required, but I'm sure we will still encounter some audio that "just can't be fixed" for quite some time. With that, let us not forget the philosophical side of the equation. As our tools advance, we need to be sure we don't use them just because we can, but because the material is calling for it. Actors' performances have life and the environment and natural acoustics of a space have attributes that create dimension and have a breath of their own. We don't want to erase that through heavy-handed processing. So often in regard to noise reduction, less is more. Our number one objective is to present the material faithfully in order to tell the best story possible.