Ratings

The ratings are based on a combination of the results from the analysis (using Earwhacks) and from listening. A good listening test can carry a bad technical analysis and vice versa, to a point. Ultimately the ratings are all subjective. I try to pick a representative song from each album to benchmark. The ratings do not reflect how good or bad the song itself is.

I'm not looking for one correct sound, and it's not about having the most expensive equipment. Indeed, unsigned artists can often produce better results then many big acts. The recordings should be appropriate for the musical style. So it's fine to have something that sounds like it was recorded in a someone's bedroom, if that suits the musical style. I'll mention "weak but appropriate recordings" in the comments on each page.

Equally you can have "three folky white guys playing skiffle in a dry sounding wood room" if the result has loads of intimacy. Allowances will also be made for age as something recorded on tape thirty or more years ago isn't going to have quite the cleanness of presentation of a modern recording. However, many of these older recordings are noticeably better than modern ones in absolute terms just through the lack of heavy compression.

Of course modern super clean recordings with inky black backgrounds are welcome too. The very best recordings pick up details like the slight variances of breathe going a instrument.

Ratings summary
	Truly awful recording. Will be hard to listen to for any length of time.
	Normally a serious technical problem, i.e. excess clipping and/or compression.
	Medium technical issues or mediocre listening performance.
	Solid recording, may have some technical issues but a good listening test or alternatively a good technical side with a decent but not outstanding listening test.
	Zero or minor technical issues with an excellent listening test.

Crimes

Compression

Compression in this sense relates not to the size of the files (e.g. compressing from CD to mp3) but dynamic range compression, making the "quiet" bits almost as loud as the loud bits. This ensures that no record ever commits the cardinal sin of sounding quieter than the one before it on the radio (oh noes!).

The recent trend for all recordings to be massively compressed has become known as the Loudness wars. See Turn me up for more information. Over compressed music will typically be lifeless and flat sounding, drums will impact and snap. Compressed music can be very tiring to listen to, much the same as being shouted at continuously. Sometimes the compression is applied at maximum level throughout the song so that the big ending just doesn't hit like it should.

Clipping

Digital music is represented by a sequence of numbers. For CDs each number is in the range of approx +/- 32000. The mastering should take the studio recording and level it so that the loudest part of the recording is just under this range. Going beyond the range means the recording the waveforms are chopped off or "clipped".

Reading the charts

Waveform

The waveform charts plot the values that make up the recording as a graph. The largest (furthest from zero) value in each x sample is used.

The waveforms give an overview of the shape of the recording. On all music types you'd expect to see a variety of large and small lines. Loud recordings (either through levelling or compression) recordings will have lines reaching to the maximum for much of the song.

Since the "full" chart is a constant width it's biased against longer songs. For example, if a song has a loud drum kick every 10 seconds. With a short song the x scale will allow the chart to show the difference between the loud kick and the surrounding quiet sections. With a longer song the kick could appear in every sample set making it look like the recording was maxed out the entire time.

The "time peaks" chart compensates for this by considering the recording in two second segements and finding the peak within each. The line drawn is the average for all the peaks covered by the line.

Clipping

The analysis tool detects clipping as samples within a small percentage of the largest value. By definition all songs will have some reading in the clipping analysis, even if it's a completely legitimate single transient.

The 1:1 chart shows a zoomed in view of the point with the most clipping candidates. In this view the x axis is per sample and the y axis is scaled one-to-one on the +/- 32000 scale.

The key points to look for, is the clipping more than one sample long? And are all the readings at that same level? If so, it probably was clipping as the true waveform disappears beyond the recordable scale. Odd shapes such as a number of samples slowing descending in value imply that some part of the software used in recording detected clipping and instead of leaving it as a flat line it wound down the values to make the clipping less audible.

The clipping candidates percentage shows how many samples were considered as possibly clipping. Any value greater than 0% is a bit poor but above 0.01% is a problem.

The "region" chart shows the clipping in context of the surrounding waveform.

Amplitude distribution

The amplitude distribution chart shows the frequency (in statistics terms) distribution of the sample magnitudes. This shows up a couple of problems in recordings.

Firstly compression is shown by curve being biased towards the right. The mode average should be at least -15dB for most songs (lower is better). Heavily compressed songs may have a mode of around -10dB. Also the curve on the right hand side may appear as bulbous after the mode point or even as a straight line.

Bad levelling is also shown. The right hand curve should reach zero on the y axis before the end of the chart. If not, you'll have clipping. The amount of the curve still above zero gives you an idea how much. Also, the clipped samples will end up in the reading for the final line giving a small spike as the final reading (this will be right against the border so can be easy to miss). In particularly bad recordings the final reading can become the mode average.

Terrorvision test

A useful benchmark for measuring a songs performance is "Alice, what's the Matter?" by Terrorvision. This is a full-on rock song yet the analysis shows a recording not as aggressive as many others. So when you've got a pleasant indie ditty with a far harsher set of charts than Terrorvision, you know something's up.