Two microphones and a comb filter

Two microphones, one source, and a sound that gets worse

Add a second microphone to a voice expecting more of it, and it frequently sounds thinner, hollow, or strangely filtered. Nothing is broken. You have built a comb filter.

Sound travels at about 343 metres per second, so it takes time to reach each microphone. If one is 34cm further from the mouth than the other, its signal arrives about 1 millisecond later. When the two are mixed, some frequencies add and some cancel.

The frequencies that cancel are those for which the delay is half a wavelength, or an odd multiple of it:

first cancellation  f = c / (2 × path difference)
                      = 343 / (2 × 0.34)  ≈  504 Hz

Then again at three times that, five times, and so on: 504Hz, 1512Hz, 2520Hz. Plot the response and it looks like the teeth of a comb, which is where the name comes from. Those notches sit squarely in the range that carries speech intelligibility.

A single microphone has no path difference and no comb. That is why one well-placed microphone usually beats two badly-placed ones.

The 3:1 rule

The standard working rule for multiple microphones on separate sources: the distance between two microphones should be at least three times the distance from each microphone to its own source.

The reason is arithmetic rather than tradition. Three times the distance means the leaked sound arrives about 9.5dB quieter than the intended sound in that microphone. At that level, the comb filtering it causes when the channels are mixed is shallow enough to be inaudible.

Two people each 20cm from their own microphone should have their microphones at least 60cm apart. In practice this means: closer to your own microphone lets the microphones be closer together, which is convenient rather than restrictive.

Polarity is not phase, and people say phase

The button labelled Ø or “phase invert” flips the polarity: every positive sample becomes negative. It applies equally at all frequencies.

Phase differences caused by distance are frequency-dependent, because a fixed time delay is a different fraction of a wavelength at each frequency.

So flipping polarity does not fix a timing difference. It is the right tool in exactly one common case: two microphones on opposite sides of the same source (a snare drum top and bottom, a guitar amp front and back) where the diaphragms move in opposite directions on the same pressure wave. There, one is inverted and the flip is correct.

For a timing difference, the fix is to move a microphone or to nudge one track in time until the waveforms align. Every editor lets you slide a clip by samples; zoom in far enough to see the individual cycles and line up a transient.

Check it in mono, always

Comb filtering can hide in stereo. Panned apart, the two signals reach your ears separately and the cancellations are much less obvious. Sum to mono and they collapse into the same signal and the notches appear in full.

This matters because mono playback is not an edge case. A phone’s single speaker, a laptop, a smart speaker, a Bluetooth speaker, many television soundbars in their default mode, and most social video on a phone: all mono. If your voice disappears in mono, it disappears for a large share of your audience.

Press the mono button on every mix, before delivery. It takes one click and catches an entire class of problem.

Stereo techniques that are designed not to comb

• Coincident pairs (XY, Blumlein, mid-side) put both capsules at effectively the same point, so there is no path difference and no comb filtering. They are perfectly mono-compatible and give a narrower stereo image.
• Spaced pairs (A/B) deliberately use time differences to create width. They are wide and they are not mono-safe.
• Mid-side deserves a mention because it is free width with guaranteed mono compatibility: one forward-facing microphone and one figure-of-eight sideways, decoded as mid+side and mid−side. Summing to mono cancels the side component exactly and leaves the mid microphone alone.

For a voice with a picture, you almost never want stereo at all. Dialogue should be mono, centred. A stereo voice moves when the camera cuts and smears in mono playback.

The lavalier-and-boom case

The common real situation: a lapel microphone and a boom on the same speaker, deliberately, for redundancy and for different textures.

They will comb if mixed at equal level. The standard approaches: use one as the primary and the other only where the first fails, cutting between them; or align them in time first (slide the boom track earlier by the path difference, which you can measure by looking at the waveforms) and then mix.

A useful measurement: sound covers roughly 34cm per millisecond. If the boom is a metre away and the lav is 20cm away, the boom is about 2.3ms late. Slide it, then mix.

The free path

Audacity, Ardour and Reaper all let you zoom to the sample and nudge a clip. ffmpeg can delay a channel precisely:

ffmpeg -i boom.wav -af "adelay=0|0,atrim=start=0.0023" aligned.wav

Most editors also include an automatic alignment function that cross-correlates two tracks and lines them up; Resolve, Reaper and Audacity’s alignment tools all do it free. For checking, any editor’s mono-sum or a phase correlation meter (Ardour and Resolve both have one free) shows you whether two channels are fighting.

The limitation

Comb filtering from room reflections cannot be fixed this way. Every hard surface creates a delayed copy of your voice arriving at the microphone, so a desk 40cm below the microphone is producing a comb at about 430Hz whether you like it or not. You cannot time-align a reflection.

The only remedies are the ones from the room lesson: get closer so the direct sound dominates, absorb the surface, or angle it. This is the concrete reason a microphone on a bare desk sounds hollow, and the reason a cloth under the microphone measurably helps.