Technical points

Frequency bands

The frequencies of many of the sources we discuss on these pages (and of LISA's best sensitivity) are far lower than that of the human ear. (The peak sensitivity of LIGO, by contrast, corresponds almost exactly to human audio.) The duration of the waves (months to years in some cases) also far exceeds what we imagine you are willing to sit through. Accordingly, we had to fudge things a bit: In these cases, frequencies are shifted by a factor of a few thousand from the way that nature would actually present them. Think of it as the audio equivalent of a "false color" image.

File formats

Sound files are generally available in .wav or .mp3 format; in some cases only one format may be available. There is no profound reason for this; generally, we make .wav files and then convert. Sometimes we haven't.

An interesting computational point

For some of these sounds, generating the data that goes into these waves is very CPU intensive. To speed things along, we solve our equations on a discrete grid of orbits and then use interpolation techniques to get data at arbitrary points off the grid. When we first coded things up, we used a simple linear interpolation to get the "off grid" data. The sound that resulted tended to be like this:


Notice that this sounds very jagged: You can hear the moments at which the small body crosses a grid point. This is clearly an artifact of the linear interpolator, which introduces discontinuities in derivatives of the interpolated data. To clean things up, we next tried a cubic spline interpolation:


The jaggedness is gone! What's fascinating is that if you were to plot the two waves on top of one another, you would have a very hard time telling which one was spline interpolated and which was linearly interpolated. Our ear is really good at picking out subtle, phase sensitive information which can be hard to see in the underlying waveform.