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

[herbstech]

One issue is when testing different audio bitrates on a laptop computer, how can I determine if my audio hardware and software are skewing the results? For example, if my audio hardware/software combo is not very good, it may make 128 Kbps MP3s and lossless WMAs sounds more alike then they really are. That would then be an unfair test. I would need to hear lossless WMA in its full glory as a standard to compare other formats to.

So without having to purchase expensive test equipment, how can one do this?

[JohnE.]

I found this posted on an audio forum a few years ago. It may not be the specific information you're looking for but I think it's relevant to any discussion involving CBR vs VBR.

"Most people do not realize CBR uses "padded samples". Every sample gets encoded at the chosen bitrate whether it needs it or not by "padding" any samples that only required, for example 167 kbps to accurately represent the audio signal, with zeros. In the case of 320 CBR, the most complex parts of the signal get 320 kbps (even if 224 or 360 kbps was optimal for the particular passage) while the less complex parts get varying bitrates based on the ABR and are "padded" with zeros to make it a constant 320 kbps. In other words, CBR is actually VBR with a quality cap at the chosen CBR and a lot of useless filler added to make the bitrate constant. Complex passages have fidelity up to the CBR cutoff point, less complex passages don't sound any better or worse than the ABR... well, that's more or less what VBR does, but without the quality cap and useless filler and VBR can encode complex passages as high as 320 kbps and beyond, if needed (unlike 320 CBR). So, why not use VBR with an ABR of 320, you ask? Simple: total overkill, defeating the purpose of using the mp3 format to make smaller file sizes; may as well use FLAC.

From wikipedia: "CBR is useful for streaming multimedia content" and "CBR would not be the optimal choice for storage as it would not allocate enough data for complex sections (resulting in degraded quality) while wasting data on simple sections."

Assuming the majority of users are storing their mp3's for playback, want good quality sound, and are not running a web based radio station (or otherwise streaming their mp3's) CBR 320 is a humongous waste of disc space, can never sound better than 320 kbps on the most complex passages and will waste data on simple passages only requiring, say 128 or 244 kbps."

Constant bitrate

Variable bitrate

My own tests with the LAME encoder have confirmed that VBR will indeed allocate more than 320 kbps for some complex sections. I've been using nothing but VBR for mp3 encoding after reading this post and those Wiki pages.

[herbstech]

"Most people do not realize CBR uses "padded samples". Whoever said that is making an assumption. How would they know that? The name implies that it is wasteful so I would think the average person that knows anything about computers and file formats would at least have an inkling.

As far as being wasteful, that is becoming less of an issue with very large capacity hard disks, jump drives, DVDs... For example, a regular audio CD can hold about 80 minutes of music max which is about 800MB worth of music encoded at 1411.2 Kbps. I typical single layer DVD is roughly 6 times that at about 4.7 GB. Lossless WMA can compress typical audio at about 2:1 with no loss of quality. If you multiply 6*2 that is 12 so a DVD with lossless WMA files can hold about 12 times as much as an audio CD or about 1000 minutes of music. Why would anyone need more than 16.5 hours of music on a single DVD? Maybe driving across the USA and are too lazy to change the DVD?

I agree compression has it's place but for audio you have to be careful. There appears to be a psychological effect if the music sounds non optimal to me. That is, I am distracted by artifacts and dont enjoy the music as much. For some "lazy" ears, 128 Kbps CBR might be good enough but many people prefer more. The 10:1 compression ratio seems to be the practical cutoff for both audio compression and image compression (such as JPGs). More than about 10:1 and the quality suffers noticeably. Note that for audio, 128 Kbps is about 11:1 compression and slightly over the "limit".

One thing I did notice is that high frequencies tend to suffer more than low freqs when you lower the bitrate too much. I am wondering why they dont just divide the music spectrum up into 3 bands, allocate a low amount of bits to that band, a medium amount of bits to the midrange band, and a high amount of bits to the uppermost band. Then somehow have the decoder put it all back together smoothly. Maybe that way a lower bitrate would better preserve the treble.

Actually our ears are very discriminating and "picky" to midrange frequenices so it is argueable which band should get the most "attention".

It is interesting to rip the same CD track to lossless WMA and 128 Kbps CBR and see how long it takes (using some good headphones or speakers) to tell which is which correctly. If you pass that test with flying colors, then try 128 Kbps VBR and see if you can do as well. Maybe have someone else play the tracks for you while you dont look (and thus dont cheat).

For some songs it may be very difficult to tell but others may be easier. I guess the trick would be to find the lowest bitrate where you can honestly not tell (you get it wrong 50% of the time), then that wouldbe the "winner". My guess (and it is only a guess), is for most people that may be around 256 Kbps CBR and slightly lower (maybe 192) VBR.

I'll try testing myself on a few different genres. I think any track with delicate high pitched sounds (wind chimes, cymbal crashes...) will be easier to tell. Not sure which sounds (or genres) work best with low bitrates.

[bigBonehead]

Herb... The compression algorithms for different formats are not comparable... WMA, MP3, OGG all depend on different schemes. In MP3 VBR, the high/low rate plus the Quality level all play into the equation... The original content of the file being compressed matters as well. A file compressed using MP3 VBR with the high rate set to 256 bps, and the low set to 64 BPS using a quality setting of 3 may yeild an average of 128 Bps, but will sound substantially better than CBR rate of 128 Bps of the same file...

This is also relative to how you listen to the file as well...

BTW.... Windows Media Player is not a good comparison tool. It colours the sound. If you want a free player that is as clean as I've found in years, try this: Wavosaur

[herbstech]

Sure they are comparable if someone is specifically doing a test to compare them. Of course they use different compression algorithms otherwise the files they output would be the same. I just listened to a 128 Kbps CBR vs a lossless WMA and it is easier to listen to the lossless WMA. The 128 has less treble and lower quality treble. It sounds like I could listen to the lossless much longer than the 128 without fatigue. It is kinda like comparing fresh squeezed orange juice with a very good reconstituted brand. They seem close but the better one makes me want to drink more.

Also, if you set a lower limit on VBR then that is not really truely variable (it is variable with restrictions or boundaries). For example, if there is dead silence during a track or a constant tone (such as a single note on a keyboard), it may "lock" at 64 Kbps or whatever the lower limit is and that is not varying. To me, VBR means it is free to go from almost 0 to some upper limit (as high as 1411.2 Kbps for 16 bit stereo (2 channel) audio CDs at 44.1 Khz sample rate). It would be interesting to take a 4 minute long dead silent track and run it thru a lossless WMA converter and see what filesize it spits out.

Audio compression is a very interesting debate. I think a good way to tell 128 vs something better is to listen to the entire song and evaluate how you feel at the end, not necessarily what you hear. If you feel "so-so" at the end, then the bitrate was probably too low. If you feel it sounded "effortless", then the bitrate was probably high enough. Kinda like a really good quality HDTV signal. It might be a crappy program that you are not interested in watching, but the very high quality of the picture lures you in and makes you watch much longer that you normally would. It also may make you overlook the content of the show and just admire the picture quality. Conversely, a poor HDTV signal takes away from the content because it is distracting to the viewer.

[JohnE.]

""Most people do not realize CBR uses "padded samples". Whoever said that is making an assumption."

Directly from the Wikipedia CBR page: "Most coding schemes such as Huffman coding or run-length encoding produce variable-length codes, making perfect CBR difficult to achieve. This is partly solved by varying the quantization (quality), and fully solved by the use of padding. (However, CBR is implied in a simple scheme like reducing all 16-bit audio samples to 8 bits.)"

[herbstech]

Quote:

Originally Posted by JohnE.

""Most people do not realize CBR uses "padded samples". Whoever said that is making an assumption."

Directly from the Wikipedia CBR page: "Most coding schemes such as Huffman coding or run-length encoding produce variable-length codes, making perfect CBR difficult to achieve. This is partly solved by varying the quantization (quality), and fully solved by the use of padding. (However, CBR is implied in a simple scheme like reducing all 16-bit audio samples to 8 bits.)"

The assumption part is that "most people" do not realize this. How would you or anyone else know what "most people" know regarding this? Did someone take a survey and ask them? Also, it depends what CBR you are talking about. A "normal" CD uses 16 bit 44.1 Khz sampling for 2 channel stereo which is 1411.2 Kbps. Where is the padding in that? That is a coding scheme but WITHOUT padding. In a way it is compressed too because it is not an optimal representation of the music on it. Recording studios use 24 bit with higher sampling rates so 16 bit 44.1 Khz sampling is actually a downmix to them and can be correctly considered compressed since some musical information is lost in the downmix (even though the lost information may be difficult for the average person to hear).

Wikipedia's wording is not very good in this case. Again they use the term "most" in "most coding schemes..." but how would they know that? Actually they dont because they dont know exactly how many coding schemes are out there thus they dont know how many "most" is.

[bigBonehead]

Maybe this will shed some light on things: Average Bitrate - Hydrogenaudio Knowledgebase

[herbstech]

It seems like all this technology to compress music files is becoming less needed because of the low cost of storage. To me, with large capacity hard disks, having a lossless copy of CDs is very reasonable. A 250GB HD can hold about 1000 losslessly compressed CDs at about 250MB each.

Audio compression is still very useful for emailing someone a song, iPods, MP3 players, and/or computers with more than 1000 songs and/or small capacity hard disks. I even tried 32 Kbps for a song and thru my "dinky" computer speakers it was at least decent. 1 MB for a 4 min song is pretty impressive. However, with headphones, 32 Kbps sounded awful compared to lossless. If I had to put a % quality on them I would estimate this assuming lossless is 100% of the original CD quality (which mathematically it is). All of these are CBR except lossless:

32 Kbps : 40% (heavy flanging effect, pretty bad with quality headphones)
128 Kbps : 80% (loss of some treble and treble quality, somewhat fatiguing)
320 Kbps : 90% (very close to lossless but treble not as "effortless")
705 Kbps : 100% (705 Kbps is about average for lossless WMA I have seen).

Someone might say this table is crap because I am cheating using VBR for lossless and CBR for the others. Well, even if I used VBR 128 Kbps, it would be no better than 320 Kbps CBR so this table has validity. VBR is kinda strange anyway because if you put an upper limit on it (say 320 Kbps), why then wouldn't it use that almost all the time (except for very quiet and/or simple passages)? The problem with compression schemes (especially for music) is they make assumptions and dont look at the entire input information first before compressing. If they did, they could do more analysis and likely get better results.

Of course these percentages will vary by person and playback hardware plus other factors.

To me it seems that CDs were a compromise and are actually not as good as they could have been (no technology ever is). So compressing them further is making things worse (unless your ears/hardware/software...) are so bad you cannot hear any difference. Just remember, if you ever upgrade your equipment and only have 128 Kbps music, then you are limited to about 80% CD quality and may have to re-rip your CDs. Why not just make a lossless copy when you first get the CD and then dork around with lower bitrates for iPods, email, MP3 players..., keeping the original lossless files intact?

It is well known that older people have a narrower bandwidth of hearing. For example, a child may hear from 20-20,000 Hz fairly well. Someone middle aged might only hear from 20 to 14,000 Hz well. For that middle aged person, they could (in theory), effectively low pass the audio they want at 14,000 Hz by lowering the sampling frequency to around 32,000 (or even 28,000) before compressing. I dont know if this will "screw up" modern compression algorithms or help them. There is even a chance that low pass filtering this way might improve the sound since it would relieve the playback hardware/software from having to reproduce sounds that the listener can't hear anyway. It may also be that those bits that would have normally been used for freqs in the 14,000-20,000 range could then be used elsewhere in the audio band where they are actually useful.

[bigBonehead]

Why do your recordings sound like ass? - Page 13 - Cockos Confederated Forums

4th post on down... WMA is, IMO, the worst lossless format... And WMP the worst player. If you want compressed lossless files I recommend you try .flac