Does it matter how we rip audio CDs?

This isn't an entirely irrelevant question in 2022. Sales of audio CDs are increasing, for the first time in 20 years. I've been buying CDs since the 80s, and I have a huge number of them -- probably many people of my generation do. Back in the dinosaur days, I ripped my audio CDs to MP3 files, because storage was too expensive to do anything else. Most portable music players only worked with MP3, anyway. I accepted the loss of audio quality this entailed, because there was no alternative. However, I've recently been re-ripping my CDs to more modern, lossless file formats, and I've been wondering whether there's any truth to the arguments that one way of ripping CDs is superior to another. After all, bits are bits, right?

Everything to do with audio is subject to strong opinions, often ill-informed. The hi-fi world has always been associated with a headlong rush to waste money on technologies that have little objective merit. The absurdities of "HiFi mains cables" have been a source of amusement to anybody with any knowledge of electronic engineering for decades. More recently, we've had to contend with the hyperbole (and waste) associated with 32-bit digital-to-analogue converters and such-like. On the one hand, the idea that one CD-ripping application will produce different results from another seems like a similar exercise in futility. On the other, if it's going to take three months to re-rip my entire CD collection, I certainly want to do it as well as possible; so there's a nagging suspicion in the back of my mind that it actually matters how it's done.

If you look at web forums for audio and hi-fi -- and I certainly wouldn't recommend doing so -- you'll see many arguments that program X is a better CD ripper than program Y. Sometimes these arguments are coherent, often they aren't. Some people use flowery "audiophile" language to describe how one program produces better results than another: program X might produce an audio file with more "body" or "presence" or similar silliness. Other writers describe the differences in terms of things that are at least measurable, like frequency response and signal-to-noise ratio.

I need to be clear here: I'm asking whether it's possible to make an exact copy of the audio data on a CD and, if so, what software exists which can do that. The impact of different compression methods and file types is important, but not relevant at this early stage. Some CD rippers can read meta-data from on-line CD databases, which can certainly be useful in some situations. But here I'm talking only about the raw extraction of audio data from the CD. And, bits being bits, it really shouldn't matter how this is done.

Or should it?

Here's the problem: the audio CD format is ancient. It was developed at a time when the basic task of reading data from the CD and passing it to a digital-to-analogue converter, at audio sampling rates, was a significant engineering challenge. The audio data on a CD is actually laid out somewhat like a vinyl record: it's a continuous spiral track, starting at the edge and proceeding to the centre. Only minimal meta-data is stored, because the designers needed to maximize the recording surface available for audio. In particular, the error checking and correction mechanism is far weaker than one that would be used for a data disk. That's why a conventional 74-minute CD can store 747 Mb of audio data but, when used as a CD-ROM, the same disk can only store 650 Mb of data. Error detection and correction was deemed to be far less important for audio use, compared to data. After all, the sound quality of an audio CD didn't have to be perfect -- it just had to be better than vinyl, for a given cost.

This is why it might make a difference how you rip a CD. If you really care about sound quality, it does potentially make a difference if your CD rip is full of errors. But how likely is that?

The "Red Book" specification of audio CD distinguishes (very broadly) between two types of error: "C1" errors are generally perfectly correctable, using information on the CD itself. After all, there is a certain amount of redundancy in the data encoding process. "C2" errors are those which cannot be corrected with complete accuracy. Provided that the CD drive is working correctly, we have no reason to be concerned about C1 errors. In practice, the reading process might generate hundreds of C1 errors per second, but all will be corrected.

C2 errors are more troublesome. Not only can they not be corrected with complete accuracy, they might not even be detectable by a computer. When an audio CD transport encounters an uncorrectable error, it will adopt a "concealment" strategy. Early CD players simply muted the erroneous data -- this was better than loud random noise. The result was a "click" of the type we got from a dusty vinyl record. Later players did better, by interpolating samples from either side of the error. This masked errors better, such that they were only really noticeable with very careful listening, so long as the CD wasn't badly scratched.

On the whole, computer CD drives do not do any error concealment -- if this is necessary, the ripping software or the operating system will have to do it.

How common are C2 errors? This is hard to estimate but, with a new, clean CD, in a good-quality drive, we can reasonably expect that there will be none at all when ripping a single disk. What this means is that the way the disk is ripped will have no effect whatsoever on the result: the resulting files will be exact copies of the original CD (possibly plus or minus a few milliseconds of silence at the beginning of the track -- finding the start of a CD track is not an exact science). We might reasonable expect to encounter one uncorrectable error, of a few milliseconds, for every ten CDs we rip.

The rate of C2 errors will increase if the disk is dirty, and increase enormously if it is damaged. In these circumstances, it might conceivably make a difference how it is ripped. Why?

First, the CD drive might not even be able to tell the ripping software that an error has occurred -- the drive might just present the software with a few milliseconds of silence. Second, even if the error is detectable, the ripper might have no way to correct it.

CD rippers that strive to be accurate when handling suboptimal CDs have to be paranoid about the data they receive from the drive. In fact, the most popular library/utility for ripping CDs on Linux is called "CD Paranoia". A paranoid ripping strategy will rip the same data multiple times, and probably as part of differing sequences. If these multiple reads produce different results, it seems reasonable to assume that an uncorrectable error has occurred. How we fix it is not clear, but at least we should know about it. Some software will read the data multiple times and, if some number of reads produce the same data, the software assumes that these are accurate reads.

Except that it's not really as simple as that, for two reasons. First, many CD drives cache data as they read it. So if you read the same sectors, the ripping application will see the same data, even if it is erroneous. Second, even if the data is not cached, most uncorrectable errors come from physical defects in the disk. Reading the same sectors multiple times, even without caching, will probably result in the same data.

Not all CD ripping utilities cope very well with caching. There is no robust way for a utility to tell the drive not to cache, so software has to make assumptions about how caching will work. It seems reasonable that if the utility reads a large amount of data from one part of the CD, then seeks to a distant part and reads another large amount, that the second read will not be from the cache. Still, there's no way to guarantee that this is the case; and algorithms that attempt to defeat caching can be slow. Even if the cache can be defeated, there's no algorithmic way to get around the fact that the same (broken) data could be read multiple times, even without caching.

So: is the notion that one CD ripping utility is more accurate than another snake oil or not?

The rate of uncorrectable errors when reading from a new, clean audio CD is much worse than would be acceptable for a data disk: it's about one error in 10⁹ bytes, which compares badly with the one in 10¹⁵ that modern hard drives offer. We don't normally have to check for identicality when copying files on a hard disk. Still, the comparatively poor error rate offered by an audio CD is still essentially negligible. To that extent, yes -- it's snake oil. If I'm copying a new, clean CD, then I don't really care what software I use to do it, or how it is set up. I expect a perfect copy, pretty much every time.

The situation is less clear when ripping worn or damaged CDs. Here it might, just possibly, make a difference how they are ripped. It's plausible that some software can detect and correct errors better than others. However, even in these cases, it's hard to be sure than one program performs better than another -- unless the results are very different. If program X produces a file that can actually be listened to, while program Y produces nothing at all, than that's a clear win for program X.

I can see no evidence, nor any plausible explanation, for claims that one program can produce better sound quality than another, when ripping CDs that are basically playable. Uncorrectable errors have to be masked, or interpolated. They will either be inaudible, or sound absolutely ghastly -- there isn't anything in between. These errors are large and, we hope, discrete. They aren't going to lead to a reduced frequency response or dynamic range.

In conclusion: if you have to rip worn or damaged CDs, it's probably worth experimenting with ripping methods or software, to get a basically listenable recording. In other cases, it really won't make any difference.