Noise, n. A stench in the ear. Undomesticated music. The chief product and authenticating sign of civilization.
[Image: Casey Kotas, Aural Repellant.]
One of the pleasures of ephemeralizing old physical media to weightless digital form is the opportunity to transcend the condition of the source material. Almost always this involves removing noise. There are two fundamental kinds of noise; each requires a unique approach to remove it.
1. Surface Noise. Surface noise is caused by artifacts that audibly impede the progress of a turntable needle across vinyl. It is typically characterized as either "clicks and pops" (caused by damage, dust, static) or "surface crackle" (caused by degradation over time leaving groove surfaces less than perfectly smooth). Where surface noise occurs, it replaces the sound that would otherwise have been recorded at that instant.
If modeled as a series of obstructions, surface noise can be removed algorithmically using a combination
of recognizer and replacement strategy. Recognizers generally look for sharp discontinuities of a specific duration in an otherwise continuous sound wave to identify noise candidates for removal. However, if you remove an obstruction, there is no sound behind it. Therefore, replacement strategies either stitch together the two sides of the wave on either side of an obstruction, or interpolate a new waveform over the duration of an obstruction.
While there are many software products to address unwanted surface noise in a recording, none are perfect and can benefit from case-by-case user adjustment of their controls to produce the desired result. Said the author of one popular tool, "it is almost impossible to get 100% accuracy with zero false positives." As a result, sometimes you are forced to remove manually obstructions that eluded algorithmic processing.
2. Audio Noise. On the other hand, audio noise is an interference signal that mixes with and degrades desired sound. It is usually characterized by its frequency spectrum, either narrowband (focused) or broadband (wide). Audio noise can be effectively removed if you know two things: a) exactly what the noise sounds like in isolation; and b) precisely where it occurs. Uncertainty along either axis has a negative effect on the result; attempting to use noise removal techniques on uncertain inputs essentially takes out random sound, further degrading the desired signal.
At the risk of over-simplifying a complex process, audio noise removal is a two step process after dividing the frequency spectrum into thousands of individual bands. First, measure the sound level of just the noise in each of the bands, creating the so-called noise floor. Next, measure the sound level of the recording (including noise) in each of the bands. If the level is greater than or equal to the noise floor, subtract off the noise floor. This has the effect of silencing bands that are just noise and reducing the noise in the remaining bands.
It has been said that 90% of success is showing up. 90% of removing noise is knowing what kind of noise it is. In both cases, the other 10% is knowing what to do. Here are two cases studies.
Artist: Anita O'DayGenre: Jazz
Year: 1955
Consider this audio sample from a direct recording of an LP from the 1950s. There is a "scraping" sound (origin unknown) in one section. Because of its several-second duration, the unwanted sound could not be addressed as surface noise. But because there was no isolated example of just the unwanted sound, it could not be removed as audio noise. In consultation with an audio expert, a mathematical model of the scraping as audio noise was created, but it really wasn't a close enough match to effect removal. Listen to the result. Ultimately, I was not able to use this track.
Artist: Perry ComoGenre: Easy Listening
Year: 1978
On the other hand, consider this audio sample from a homemade cassette recording of an LP with noticeable wear and tear. Using a combination of techniques, three layers of audio grime could be removed: clicks and pops, surface crackle (both surface noise), and tape hiss (broadband audio noise). Listen to the result. The entire recording was generally in this degraded condition, and I was very happy to be able to restore it to the point it could be included in my Mom's iPod library.
© 2012 Thomas G. Dennehy. All rights reserved.
Why would you want to remove record noise? Its the best part! I even purposefully dip my records in a sandbox and scrape them with a butcher knife to make it sound even better!
ReplyDeleteSound like vinyl is good. Sound like new vinyl is even better.
ReplyDelete