Buzz is a word that we use to refer to many audio degradations. It sometimes describes the noises produced by, for example, problems with lighting rigs and faders. These noises often take the form of closely spaced, regular 'ticks' in the signal. On other occasions, 'buzz' is used to describe the sounds produced by electrical faults such as earthing problems. These can lead to a variety of unwanted noises such as simple mains hum or harmonically complex... well, buzzes! (Figure 1).

Figure 1: Spectral analysis of a typical harmonic buzz

You can view and modify audio signals in two principal ways - in the time domain (in which you consider the waveform of the signal), and in the frequency domain (in which you consider the spectral content of the signal). It is important to remember that these domains are just different ways of considering the same information. Nevertheless, it is more effective to handle the audio and modify it in the domain that is most appropriate to the problem. Therefore, for example, a buzz comprising closely spaced, regular 'ticks' is best restored using one of CEDAR's various decrackle (time domain) processes, and we will not consider it further here. On the other hand, a buzz that has an indistinct time domain structure may be more easily characterised by its harmonic content and then analysed and treated in the frequency domain.

Historically, audio engineers have used a number of conventional filtering techniques to eliminate buzzes and hums. The simplest of these is the basic high-pass filter, which attenuates the signal at all frequencies below a specified cut-off point. This can be used to eliminate harmonically simple hums which, because of the mains frequencies of most countries, have fundamentals of 50Hz or 60Hz. Unfortunately, such a filter will also eliminate some of the desired sound as well as the unwanted tone, resulting in a "gut-less" signal that lacks bass.

Figure 2: An High Pass Filter

Some complex buzzes and hums exhibit harmonics reaching up to many kHz. But these too have significant energy at lower frequencies, so a low-pass filter is completely inappropriate for removing buzz.

A more sophisticated approach involves the use of comb-filters - so-called because the response of the filter resembles the teeth of a comb. (Figure 3).

Figure 3: A Comb Filter

These filters allow you to remove signal components at a number of regularly-spaced frequencies. Furthermore, the tighter the 'teeth' of the comb are, the more precise the removal can be. This looks, at first sight, like an ideal solution, but it isnšt. For one thing, the filter exhibits the same behaviour as all other static filters: if, at a given frequency, you remove an offending noise, you also remove any genuine signal that may exist at the same point in the spectrum. (Figure 4). Secondly, comb-filters introduce a characteristic "boxiness" that makes the genuine signal sound distant and indistinct. This problem becomes worse the more precise the 'teeth' become so, in limiting the unwanted side effects caused by the first problem, you exacerbate the second. Thirdly, the offending frequencies may not be entirely constant, so the filteršs 'teeth' may not always be appropriately placed, and you will hear the buzz come and go during the course of the material you are playing.

Figure 4: Spectral analysis of a comb-filtered signal

These problems make comb-filters - which are, nonetheless, implemented on a number of audio processing platforms - unacceptable for the highest quality removal of buzzes and hum. A far better solution is an algorithm that identifies the offending frequencies, tracks any fluctuations that they experience, and then removes the noise without destroying the underlying signal. (Figure 5). CEDAR's Debuzz process precisely fulfils this specification. It will track buzz frequencies within a range of ą2%, and will retain genuine signal components while removing any buzz harmonics that exist at the same frequencies.

Figure 5: Spectral analysis of a signal retored by the CEDAR Debuzz process