The important difference between audio and file compression

It’s tempting to compare streaming quality based on the numbers alone, but it’s wrong. If you’re looking to make a decision on streaming services, you’ll want to check what codec the streams use, rather than what the bitrate they stream at is. While that flies in the face of common perception that a larger number means better quality, it’s an important concept to understand.

The problem

Audio compression is a big subject, ranging from studio editing to converting file types, and even sending music over Bluetooth. The word “compression” might throw you for a loop because there are so many ways it pops up in discussions about audio quality that it can be tough to follow—and in fact, many people make that exact mistake when talking about streaming quality. It definitely can stir a negative connotation for anyone familiar with the “loudness wars,” but digital compression codecs like the MP3 are the cornerstone of modern online streaming services.

Digital file compression codecs don’t alter the mix or perceived loudness like dynamic compression in studio mixes does. Instead, file compression uses clever algorithms to shrink down an uncompressed 50MB or so music file into something much more portable, say around 7MB or smaller. That’s obviously great news for your data plan if you stream a lot. That being said, audio file compression can be the difference between a muddled mess and toe tapping bliss. We’ve all heard our share of terrible MP3 rips and flawless FLACs. Although music compression codecs are mostly very good these days, there is still plenty of misinformation and confusion out there about what makes for good quality experience.

How audio file compression works

Let’s start off dispelling the big one, fewer kbps (kilobits per second) doesn’t always sound worse. By the same token, more kbps doesn’t always sound better. This assumption is often made by incorrectly associating audio with regular files. To make a standard digital file smaller implies that you’ve squeezed something down or chopped more out. Obviously audio files can work like that too, but modern compression is so much smarter.

Digital audio compression comes in a growing number of forms, from the files streamed from online services to the data sent to your Bluetooth headphones. For starters, almost all audio compression codecs are lossy—as opposed to lossless—meaning that some information is removed. But that’s not strictly a bad thing for quality; provided you remove data that you can’t hear anyway. After compression, the file is decoded using the reverse codec when you want to listen.

256 kbps AAC is equivalent to 320 kbps MP3. Who knew?

File types like MP3, AAC, and OGG exploit the limits of human hearing by using psychoacoustic compression principles to remove parts of the file that we simply can’t hear in order to reduce file sizes. Encoding and decoding isn’t particularly time sensitive, and is done in the frequency domain using MDCT. These algorithms can be very accurate and produce great sounding results with very small file sizes. Think about it like meticulously picking out which sounds can and cannot be heard: It takes a while, but is very difficult to distinguish from the origination source when done properly.

Real-time audio codecs, like Bluetooth’s SBC, aptX, LDAC, and others, tend to use noise shaping as a form of compression because it’s faster and can be done with lower latency. This type of compression isn’t content specific like psychoacoustic compression is, but it still abuses the sensitivity of our ears to make compression optimizations. On the whole, results are quite good—but these algorithms have to complete very quickly. They therefore can’t squeeze file sizes down quite as far as psychoacoustic compression types do for the same given audio quality.

There’s also some overlap between the two, depending on your latency and processing requirements. AAC, for example, is used as both a file and Bluetooth codec. Both of these types of compression can also be taken to the extreme, producing noticeable compression artifacts. However, the file sizes from streaming services and over Bluetooth are more than good enough to avoid these problems.

This is what a 24-bit music file looks like before any data is removed. Frequency is the Y-axis, time is the X axis, and intensity is color.

Finally, there’s also completely lossless file types like FLAC. These types exploit coding rather than auditory optimizations and certainly don’t compress files to anywhere near as small. Despite the larger file size, it’s generally considered impossible to tell the difference between a 320 kbps MP3 and a FLAC file that’s 5x larger.

Not all bitrates are equal

With the background out of the way, it’s important to recognize that not all compression algorithms are equal. Therefore, comparing quality based on file size or bit rate (in kbps) of two different file types is putting the cart before the horse. It’s not always possible to compare audio quality by bitrate alone. If you accept the premise that different compression algorithms can offer different qualities for the same bitrates, it stands to reason that sometimes a lower bitrate file can sound better than a higher one—depending on the codec, anyway.

There are a few factors that determine whether some bitrates are better than others. First is the algorithm type and what it’s designed to do, as we outlined previously. Non real-time, psychoacoustic codecs can actually be much smaller and sound better than low-latency time codecs. Next is the quality of the algorithm and the encoder used (AAC is an improvement over MP3, for example), and Opus is considered better still for any given file size. Back in the early days, different MP3 encoders produced different sounding results.

…comparing quality based on file size or bit rate (in kbps) of two different file types is putting the cart before the horse.

So how do we measure quality when the numbers don’t tell us everything? Well, we look to a more subjective test called the ITU-R BS.1116-3 perception test. Essentially, people listen to several files in both lossless and lossy versions, and then rate the difference on a scale of 1-5. A score of 5 is imperceptibly different than a lossless source, and a score of 1 is a very annoying degradation of sound quality. Even though using humans as testing equipment isn’t exactly scientific, it can provide some context for us when we’re trying to compare file formats and their perceived performance.

For our purposes, check out the results from this 2014 Public Multiformat Listening Test. The test compared listener responses to five different popular compression codecs set at approximately 96 kbps, ranking them each for quality and distortion audibility.

listening-test.coresv

Despite the very similar bit rates, the Opus codec scored consistently higher than it’s competitors, followed by AAC and Ogg. The MP3 came in fourth despite often encoding at a slightly higher bit rate. FAAC consistently performed the worst, highlighting that there’s much more than just bit rate to determine audio quality. Also note how that even at 96 kbps, heavy compression from the four most popular codecs didn’t rank as annoyingly bad.

Bluetooth LC3 scores higher for sound quality than SBC and also runs at lower bitrates.

More recently, the LC3 Bluetooth codec at 160 kkps claims to sound better than SBC at 345 kbps. We’ve conducted our own tests too and found that the aptX Bluetooth codec also beats SBC for sound quality, despite a very similar 352 kbps bitrate. Likewise, aptX also sounds better than LDAC’s 330 kbps setting, but it’s a much tougher comparison when you compare LDAC 660 kbps and aptX HD at 576 kbps.

The key takeaway

Enough of the hypothetical, what about the real world? Well, with this newfound knowledge we’re now in a position to look at digital download and streaming service with more of an eye for quality. Let’s quickly rundown the quality on offer from the most popular streaming platforms.

See also:

Best music streaming services

As you can see, there’s a surprising similarity between most of the streaming services. AAC is the most popular type of lossy compression, while Hi-Res libraries offer FLAC as standard. Providing you have a decent enough internet connection, you’re unlikely to notice any perceivable difference between the vast majority of these services. At around 192kbps, AAC is a very good sounding codec. Anything above that is a bonus.

The exception to the list is Spotify, which offers 96 kbps as its “normal” setting, which is not as high as some of its competitors. Although based on the research we highlighted earlier, 96kbps AAC actually sounds pretty good. Fortunately you can manually set this to high or obtain very high’s 320 kbps setting if you’re willing to pay for it.

Bluetooth codecs have higher bit rates for a given quality compared to streaming file types.

Most music streaming services now offer very similar levels of music quality. While you can pay more for hi-res if you deem it worthy, the lossy compression options out there are virtually indistinguishable from CD quality. So the next time you hear someone complaining about music streaming quality remind them it’s much more likely to be their cheap Alexa speaker or headphones that sucks. Or perhaps their internet connection just dropped out?

Want your music streams to sound great? Check out our picks for best headphones.