Technical notes index

Principles of
            audio noise reduction
Principles of audio noise reduction
<<previous page       page 4       next page>>


      Frequency response
      Pitch errors






Random noise (hiss)

Rumble can also be regarded as random noise, and much the same considerations apply to its removal as hiss, which is what will be considered here.

Whereas distortion, wow and flutter are almost impossible to remove, it's possible to effect a considerable improvement in random noise levels. However, any such process tends to affect the original sound, so there is always a balance of judgement to be made - reducing the noise against taking the top off the recording. Unfortunately many transfer engineers seem intent on removing almost all the noise, apparently unconcerned about the woolly sound which results.

Obviously, the simplest way - and for many years the only way - to remove hiss was to filter it. Of course this had serious effects on the actual recording, and it became normal to expect 78rpm records to sound like this. With the development of more sophisticated methods it became possible to have far less effect on the original sound. There were a few analogue systems, most notably the Packburn (now available in a new version), which plays 78s with a stereo pickup and continuously compares the two groove walls, switching rapidly between them to choose the quieter to remove clicks and plops, as well as providing hiss reduction and equalization. The first digital system, CEDAR, was developed in 1988 after several years research. Initially it took all night to process one 78rpm side, but now it's available both as software and as hardware, which can process a record as it's played. However, it's extremely expensive. Other systems include Sonic Solutions' 'No-Noise' at a professional level (now superceded), and in more recent times inexpensive systems such as Bias SoundSoap (no longer available) and Adobe Audition have become available and work on home computers: I use ClickRepair and DeNoise by Brian Davies (unfortunately now unavailable).

The basic process for dealing with hiss involves splitting the audio spectrum into a large number of very narrow frequency bands - from 256 up to some thousands - and working on each one separately. A sample of the recording where only the noise is audible can be used, or a pre-determined template, and for each of the frequency bands a level is noted which represents the noise.

 In the case of each narrow band, any audio which is lower than the determined threshold is successively reduced in level as the level drops - so that a 3 dB drop in the original might be replaced by a 6dB drop: this is done until a level of reduction is reached which can be set in the program. Above the threshold the audio is unaffected, in the hope that this is the original recording and not noise. (Many 78s have more noise than recording at higher frequencies and this presents a considerable challenge to hiss-reduction programs.)

There are a number of parameters which can be either manually set or more usually are preset in the program: the point at which as the level drops the reduction ceases: the speed with which the reduction is applied when the level drops and removed when the level increases: and the amount by which adjacent frequency bands affect the behaviour of each band (large differences in the levels between bands can cause unpleasant effect similar to a very low-bitrate mp3). There are other more sophisticated considerations in the design of these programs, which is why they all behave differently.

The next page includes some examples of hiss reduction.

<<previous page       page 4       next page>>

Roger Wilmut.