Audio: comb filter (0:03)
comb filter plays bird chirps. Each chirp is a sine wave with a variable frequency that curves downwards from 8kHz to 3kHz. The first four chirps are unfiltered and the last four chirps are comb filtered by adding a copy to the original unfiltered sound and delaying the copy by 30 milliseconds. The comb filter figure shows a close up of part of the spectrogram of the comb filtered sound with its distinctive regularly spaced peaks and troughs.
Comb filtering combines a sound with a slightly delayed copy.
Comb filtering is easy to write. Generate a sound, any sound, make a copy, delay the copy slightly so that it overlaps the original, then mix the delayed copy with the undelayed original. It is as simple as that.
The delay in comb filtering causes the original sound and the delayed sound to overlap. When two sound waves interfere with each other it causes the amplitude of some frequencies to increase, through constructive interference, and the amplitude of some frequencies to reduce, through destructive interference.
The delay time in comb filter is fixed. This fixed time delay generates an alternating series of equally spaced peaks and troughs in the spectrum of the output. The spectrogram looks a bit like a comb, which explains the origin of the curious name. The result is a hollower sound, due to the reduced amplitude of the frequencies which form the troughs in the spectrogram.
A comb filter with a fixed delay time generates a noticeable fundamental frequency in the output. The frequency of this fundamental is the inverse of the delay time. For example, a delay of 2ms generates a fundamental frequency of 500Hz and the peaks in the spectrogram will be equally spaced 500Hz apart as a direct result of the 2ms delay time. The equally spaced peaks form a harmonic series.
Comb filtering is similar to phasing. Both produce peaks and troughs in the spectrum. Both produce a hollowed out sound. The difference is that comb filter peaks are regularly spaced because the time delay is constant, whereas the allpass filter peaks in phasing are irregularly spaced because, whilst the phase shift is constant, the delay time varies with the frequency. In addition, comb filter peaks extend to the Nyquist frequency whereas the number of allpass filter peaks depends on the order of the filter.
Comb filtering is also similar to echo. The main difference is that comb filtered sounds overlap whereas true echoes are discrete sounds.
Comb filtering can be a problem in sound recording with a microphone. Using a single mike to record multiple sound sources will inevitably produce comb filtered sound. For example, the back row of a choir standing six feet behind the front row will produce a delay of around 6ms. A common solution, fondly known as the 3:1 rule, is to add another mike equidistant from the source and at least three times that distance from the other mike. Similarly, recording a single sound in stereo will cause problems if one mike is closer to the source than the other and the two sounds are mixed to mono. The solution is the same, space the mikes equidistant from the source and at least three times that distance from each other.
Comb filtering is a versatile and simple effect for writing music. Adding a delay in the 30-50ms range adds a feeling of space to a melody or a harmony. A longer delay time, up to 100ms, creates a comb filtered harmony with a distinct echo reminiscent of many a rock n roll record in the mid twentieth century. Comb filtering rhythm, whilst unusual, is perfectly feasible, and is especially effective when the delay time matches the tempo of the rhythm. Comb filtering sine, saw, triangle, pulse and square waves produces spacey echoey sound effects.
