128 Kbps (CBR vs VBR) vs. higher audio bitrates

[herbstech]

There is still much debate on this subject so I wanted to get some feedback from members here about their experiences with 128 Kbps CBR (Constant Bit Rate) vs. 128 Kbps VBR (Variable Bit Rate) vs, higher bitrate compressed audio formats using uncompressed CD quality 16 bit 44.1 Khz 2 channel stereo as the source audio.

I read online that 128 Kbps, (especially VBR), is adequate for many casual listeners and especially for "harsh" audio environments such as in a moving car and especially for typical computer speakers.

For high quality home stereos, many articles state 128 wont cut it. Also, people with "good" ears may find even 128 VBR not as good as the original recording.

I will be doing my own tests with a laptop computer and headphones and testing to see if I can tell the difference and how much effort it takes.

I think a good way to "cheat" is to purposely encode the original recording at a very low bitrate such as 64 or 96 Kbps CBR so you can hear the artifacting and other audio quality loss. Then just listen for similar with the 128s, 160s... That would be roughly equivalent to making a highly compressed JPG image from the original, looking for artifacts in it, then recompressing the original image multiple times progressively less and less until those same artifacts "go away".

Since one persons testing and results cannot be indicative of what everyone may experience, I am asking for other people's results here.

I have a Vista laptop so I'll be using WMA format. I'll try using 2 test files, one symphonic music and one something else like pop or rock.

That brings up another interesting related topic: Which musical genre is most likely to experience audio artifacting and/or general audio quality loss from lossy compression? I guess that answer would depend on the codec being used and it's limitations as well as a bunch of other factors.

[Whir]

Personally, I can hear how terrible 128kbps sounds on just about anything. My minimum is 192kbps CBR, but I prefer 320kbps CBR because I'm a sound nerd. Anything that I want perfect quality for, I use FLAC.

Keep in mind that VBR isn't better quality, it's actually worse. It basically relies on magic to dictate what bitrate to use for a certain section of music. The more information present, the higher the bitrate will be. It was developed to save disk space, something which most people don't need to worry about anymore.

The first thing you'll notice low bitrate artifacting in is cymbal crashes. Pop and rock especially. Past that, anything with a lot of high frequencies. Most classical doesn't have a lot of frequencies in that range that aren't upper string or vibration harmonics, so you wouldn't notice it as soon. Bass will also sound warbly, but it's much more apparent in the highs.

[RobRich]

Quote:

Originally Posted by herbstech

I have a Vista laptop so I'll be using WMA format.

If using WMA 10 Pro, 128kbps should suffice for general music listening. If using an older WMA format, be prepared to start cranking up the bitrate.

With that said, is there any particular reason you are not using AAC for lossy encodes?

[osprey4]

Using typical crappy headphones, such as those than come with an iPod, 128 kbps is fine. There's no point encoding a higher bitrate if you don't have the equipment to hear the difference.

[herbstech]

I rip my CDs using lossless WMA so I have a 100% quality backup in case I lose the CD or it gets damaged. Lossless WMA is advertised as 470-940 Kbps but I have seen at least one track even higher but it was a distorted recording to start with (yes someone messed up in the recording process). The bitrate of that was over 1000 Kbps. The general pattern I have seen so far for WMA encoding is tracks with lots of passages encode using lower bitrate and those with many loud passages use a higher bitrate.

It is interesting that they use VBR algorithms now but I dont quite see how VBR can be worse than CBR. For example, suppose we use 128 Kbps for both (on average). It seems like the VBR should ramp up the bitrate to maybe 160 for passages that need it and maybe 96 Kbps for other sections that require less. Assuming it works correctly, that should theoretically provide MORE overall quality since something like a loud cymbal crash should sound better at 160 Kbps than at 128 Kbps (instantaneous bitrate).

Also, it is important to note that when compressed, # of bits is not an absolute indicator of quality (unless you use the same exact algorithm such as some specific MP3 codec for example, and the same exact source such as uncompressed CD audio for example).

I sometimes wonder if it would be better to bust up the audio spectrum into 3 bands (bass, midrange, and treble), and encode each of those separately tweaking the algorithm for each. That is, I doubt bass and treble require similar encoding. Each band could have a highly tailored algorithm. I suppose the split points (crossover points) might be a problem since a decoder would have to blend them back together without artifacts.

Generally speaking, a compression algorithm cannot be optimal unless it first "sees" all the data it needs to compress, and even then there is no guarantee any algorithm would be optimal. I doubt this "prescan" takes place in ripping. I suppose they decided optimal compression was not worth the effort so they probably just read in chunks and compress each of those chunks.

It is my understanding that audio compression has advanced in recent years and 128 Kbps is becoming more like CD quality but from the few responses I got so far, people can still tell. Note than if and when you do a test, it is important to start with a CD you are very familiar with. Dont just download some 128 Kbps song because you dont know what they started with as an audio source for that.

For my test, I will use headphones and carefully compare a 128 Kbps CBR vs. 128 Kbps VBR vs. original lossless WMA (usually around 700 Kbps).

After I can pass that test (be able to tell which is 128 and which is not), I will bump up the bitrate to 160, then 192... until I cannot tell any difference from lossless.

Again I am using Windows Media Player that came with Vista Home Premium. That one allows MP3 output too. I think my ears are pretty good and I am a drummer so I can usually tell when a cymbal crash sounds fake. Some peoples ears are so bad, you could turn just about anything into a 128 Kbps MP3 and they wont notice. Part of that is playback on crappy equipment that distorts the sound anyway so what is a little more distortion to them? Nothing!

[RobRich]

You seem hung up on "128kbps" as some type of standard, but 128kbps can produce wildly different results depending upon the formats being used: MP3, AAC, HE-AAC, AC3, Vorbis, WMA, WMA Pro, Real Audio, etc. For example, 128kbps AAC can be considerably different than 128kbps MP3.

Ideally, you should do various encodes to whatever formats you want, then utilize ABX testing with multiple comparison iterations. Without ABX testing, you are largely wasting your time due to psychological responses involved.

[herbstech]

The purpose of my test is not to be very scientific about this using sophisticated tools, but rather to use the human element to determine if I am fooling myself into believing the "standard" 128 Kbps bitrate is "bad" and higher bitrates are much better. That is currently what I believe but based more on theory and research than on actual tests. The proof is in the pudding. That is, if I can tell the difference 90% of the time or more doing an A/B test, then I would say the statement is valid. Since I am using Vista Home Premium, I will use whatever outputs they have which I think are WMA, WMA Pro, WMA lossless, MP3... along with CBR and VBR variants.

Again I repeat, this is a test based on my ears and brain, not scientific equipment. What we perceive is what really counts, not what actually is. That is, if someone's ears are so bad that they cannot tell ANY difference between 128 Kbps WMA and the original CD music, then they have a valid point for downconverting and as a side effect, they save storage space. Another slight advantage of 128 Kbps is songs are about 1MB per min encoded so it is rather easy to predict how many 128 Kbps songs will fit on a 1GB iPod for example (4MB per 4 min song would normally be 250 songs per GB but assuming some iPod overhead, maybe 200 songs). I do own a 1 GB iPod and dont even know if it can handle lossless WMA (I doubt it), but even if it could, assuming 768 Kbps on average, I would then only get about 1/6th as many songs or about 33 of them.

Does anyone know the bitrate limit of those super tiny iPods (the ones without any display)? I think they are called iPod Shuffle. I think mine is 2nd generation. If you attempt to put a very high bitrate song on there does it automatically reduce the bitrate to something like 256 or 320 Kbps max?

[RobRich]

What you "perceive" can be influenced by your expectations of formats and bitrates, which is a common psychological response and largely the reason ABX testing is recommended.

The iPod does not natively support WMA. It does natively support AAC. I suggest using iTunes (or another AAC encoder) instead of Windows Media Player.

IIRC, the maximum AAC bitrate for the iPod is 320kbps. If you are not satisfied with lossy encoding for the iPod, then you should look into Apple Lossless.

On a related note, if you can clearly distinguish a quality difference between AAC 320kbps and CD source, then you need to be working for Fraunhofer IIS.

[herbstech]

Now I am even more anxious to try the test. I just hope my laptops built in sound hardware is good enough not to mask the differences. I have some half decent headphones made by Koss I think. Also, it is important for me (and others) to try to determine if the compressed sound is identical to the original and not to pick which one sounds better. It may be the case that compressed sound actually sounds better to some people maybe because they like that sound. For example, if by compressing some dynamic range is lost, some people might think that sounds better because the average volume might be higher and/or the quieter passages might be slightly louder.

I am very curious to purposely encode a piece of music I am very familiar with at a very low bitrate (like 64 to 96 Kbps) so I can hear artifacts easily and then know what to listen for. Some might say that is cheating but I say it is "preparing". That is, armed with that knowledge, a codec can run, but it cant hide.

I'll post any results here, give me a few days to play around with testing.

[herbstech]

I can hear a difference between 128 Kbps (CBR MP3) and lossless WMA. It is most noticeable in subtle sounds such as wind chimes and other more powerful sounds like cymbals crashes. Also the 128 CBR fatigues my ears more as it sounds a "little off". The lossless sounds very "inviting" and I want to turn it up louder.

I did also test 48 Kbps (also CBR but WMA) because that was the worst my computer software offered. It was of course very bad but it trained my ear what to listen for. The cymbals in the 48 mode sounded a little like they were mixed with metal trash can lids.

320 CBR was very good and so was 333 (VBR) and those are hard to distinguish from lossless for the short tracks that I tested (I didnt test long tracks yet).

A decent analogy to describe the difference between 128 CBR and lossless can be 16 bit color vs. 24 bit color. From a distance, they look the same but up close, you can see color banding on the 16 bit. Similarly, if you have headphones and/or very good speakers, the 128 CBR will show flaws in the sound but if you are only using it for background music or have cheap speakers/headphones, 128 CBR will suffice.

Note that the lossless bitrate was automatically selected by the ripping software and it was 857 Kbps, so it makes perfect sense that anything lower would not sound the same as the original.