CEDAR invented spectral editing and still holds the patents to it. But not content to leave things as they were, the company has continued to research ways to improve it, making it faster and easier to use, able to cure a more extensive range of problems, as well as making it even more productive in a wider range of settings. To that end, CEDAR has announced Retouch 8, which adds two new capabilities to the original spectral editor.
Matching – Applying Machine Learning to Identify Problems
There are many occasions when you will encounter similar instances of unwanted sounds in a track – things such as hi-hat spill in a multi-track recording, over-excited sibilants and plosives in a vocal track, or even the repetitive noise caused by a piece of machinery in the background. Traditionally, one would remove these by identifying each event individually and then defining it manually prior to processing. But, this is time consuming.
A new Match facility allows users to mark one of the offending sounds and then ask the machine learning algorithm in Retouch 8 to find all of the other instances within the recording. A threshold control even allows you to bias its decision making toward including more or fewer possibilities so that you can find the level that includes everything that you want to remove, but precludes events that seem similar but are actually wanted signal.
Having identified all of the matches, users can then eliminate them using the appropriate Retouch tool. You can process them individually, authorising and applying the chosen tool to each, or you can process all of the matches as a group, whereupon Retouch 8 will fly through the whole file in a tiny fraction of the time needed with human intervention.
Repair – Putting the AI Into Spectral Repair
With spectral editing, users mark a region they want to suppress or eliminate a noise within it. If interpolation doesn’t work quite as well as you want, you can’t patch it, and if you attenuate the whole region you obtain a hole in the audio that generates an audible artefact. What you need is a process that can suppress the event while leaving the background in the marked region untouched so that the resulting audio has no unwanted event and no change in the background signal and ambience.
Retouch 8’s Repair uses artificial intelligence to achieve precisely what you need. Just mark the sound and some of the surrounding signal in the usual fashion and then use the single knob to suppress it or even completely eliminate it. Unlike other spectral editing tools, only the significant signal within the region is identified and processed; all low level signals as well as the ambience are left unaffected.
Alternatively, you may have a clip in which there’s something – for example, a quiet word or some other understated sound – that you might like to bring to the foreground. Repair allows you to do this, too. Just amplify the event while leaving the rest of the signal unaffected.