In December 2018 we published a blind listening test in which we compared the results of 4 top noise reduction plug-ins across a dialog track that had a fair amount of audible hiss. The plug-ins used in this test were:
Accusonus ERA Noise Remover (Formally known as ERA-N)
Acon Digital De-Noise (Part of Acon Digital’s Restoration Suite)
iZotope Voice DeNoiser (Part of iZotope RX Bundles)
If you didn’t have the chance to take part in this test, or you would like to hear the examples again, you can do so by checking out the original article:
Of the 114 who took part in the blind listen test and voted in the poll, 35% prefer the results of Example 2 (Accusonus’ ERA Noise Remover) followed closely by Example 3 (Waves Z-Noise)
Example 1: Acon Digital De-Noise
Example 2: Accusonus ERA Noise Remover
Example 3: Waves Z-Noise
Example 4: iZotope Voice DeNoiser
What Do We Make Of These Results?
It’s worth mentioning that the four plug-ins used in this test are different in terms of their under the hood technologies. They all may reduce noise but they ways in which they all go about noise reduction is different. A plug-in developer friend of ours reached out to us and raised a good point that we felt was well worth sharing in this test conclusion. iZotope’s Voice DeNoiser works in the time domain while the other three work in the spectral domain. We didn’t mention this in the original test article, so what’s the difference between the two? Our plug-in developer friend states:
Time domain noise reduction, such as in iZotope’s Voice DeNoiser, produces fundamentally different audible results since it essentially works as a multi-band time-domain gate. A time-domain approach will allow significant amount of noise to pass when voice is present therefore it causes no voice filtering and it can sound more natural when you listen in relatively low-level conditions. On the other hand it can become really annoying in high-level conditions (for example in a cinema) where you hear a lot of noise each time the actor/actress speaks. Therefore, in this experiment the subject ends up comparing apples to oranges. IMHO this will bias the results and it will create an "anchor": if a lot of people listen in normal listening conditions (which will probably be the case), iZotope's solution would probably be the most preferred. In this case noise will be masked by the speech and the subjects will focus on the voice clarity. On the contrary if a lot of people listen in loud conditions (probably not the case here) the opposite might happen: people will focus on the fact that noise is present and will rate based on the noise suppression inconsistencies.
One thing is for sure, the time domain technology in iZotope’s Voice DeNoiser certainly didn’t bias the results in this test. What do you take away from these results? If you regularly denoise dialog audio tracks then which of these plug-ins do you use or have experience with? Personally I go between all four of these plug-ins featured in this test. If I find that one of these plug-ins mangles the sound of the track I’m working on too much I quickly switch to a different plug-in until I get the transparent results I demand. All four of these plug-ins which provide noise reduction are all essential in my workflow, so from my point of view there is no clear cut winner or loser in these results, however, it is interesting to see that the easiest noise reduction plug-in to use and set (ERA Noise Remover), is also, if only by a small margin, the most preferred in terms of sound quality.