MP3 Bit Rates, File Size and Audio Quality

In a previous blog post I talked about the need for both file and audio compression on your podcasts. At the end of that post, I linked out to an article from Scott Burton on audio compression, but did not have any information on MP3 file compression at the time.

NewImage

Well, I have all the MP3 file compression information, along with bit rates and my own judgement on audio quality vs file size now. And you might be surprised at the results! I know I was!

The Episode Content And Bit Rates

I recorded a 32 minute podcast episode that contains 2 things:

  1. A musical introduction
  2. A lengty spoken track

The musical introduction takes not quite a minute, with the spoken audio taking up the rest of the 32 minutes. The spoken audio is not a complete episode, however. It is only a 1+ minute description of the episode’s purpose, repeated over and over and over again to get a podcast episode of approximately average length. In other words, don’t bother listening past the first minute and a half. It just repeats for a long time to make the point of file size vs content length.

After recording the episode, I exported it under 10 separate bit rate settings, using Aadacity. With all of these bit rate exports, I create a special podcast on SignalLeaf just for these episodes. You can subscribe to the podcast RSS feed and listen to each of the episodes in iTunes and other podcast players, or you can use the embedded players here.

Constant Bit Rate Versions

These episodes were exported at a constant bit rate – a rate that stays the same throughout the entire encoding, regardless of the amount of data that there is to encode.

16K Constant

Episode exported at 16K Constant bit rate with 20K sampling (max supported by 16K file bit rate). The effect is quite horrible. Everything sounds muffled, warbly, and like its in a tin can with ear muffs. The file is quite small, at 3.9Megs, though.

32K Constant

Episode exported at 32K Constant bit rate with 44.1K sampling. The effect is still pretty bad. Everything sounds warbly and slightly muffled, though not as bad as the 16K file. The highs are especially off, and a lot of the stereo effect in the opening music is gone. The file is still pretty small, at 7.7Megs.

64K Constant

Episode exported at 64K Constant bit rate with 44.1K sampling. The quality drop in this file is negligible. There are some points in the opening music where some slight warbles can be heard on the high end sounds. Additional guitar sounds and other instruments tend to have a more pronounced warbled sound at this bit rate. The file size is still quite reasonable and may be considered “small” by some standards, at 15.5Megs.

128K Constant

Episode exported at 128K Constant bit rate with 44.1K sampling. The net effect of this bit rate is no noticeable degradation in audio quality, throughout the podcast episode. If additional instrumentation were present, there would be little to no loss of audio quality. The quality comes at a trade off in file size, though, with the file being 31Megs in size. This size is hardly acceptable for a podcast that is only 32 minutes in length.

256K Constant

Episode exported at 256K Constant bit rate with 44.1K sampling. At 62 megs, this file size is completely unacceptable for distributing a podcast of only 32 minutes in length. The sound quality, however, is near impeccable. There is little chance that anyone would be able to pick up any distortions in the quality of audio.

Variable Bit Rate Versions

Variable bit rate encoding allows the bit rate to change based on the amount of information that needs to be encoded. If there is less audio signal to encode, the bit rate goes down. If there is more, it goes up. This can reduce the file size while still maintaining quality, in many cases.

48K – 85K Variable

Exported at 48K-85K Variable bit rate with 44.1K sampling. There is very little distortion in this file, even in the musical introduction. A direct comparison of the 64K Constant bit rate yields a slight increase in audio quality over that export. The file size, in comparison, dropped from 15.5Megs down to 13Megs. The audio quality vs file size at this bit rate is quite nice.

80K – 120K Variable

Exported at 80K-120K Variable bit rate with 44.1K sampling. There is little to no perceivable distortion in this audio even in the musical introduction. A larger assortment of musical instruments comprising a full band or orchestra may exhibit some minor issues, but not enough to cause serious displeasure. A direct comparison of the 128K Constant bit rate yields approximately the same audio quality, though it is technically slightly less. The file size, in comparison, dropped from 31Megs down to 18.4Megs for a mostly spoken audio file. The audio quality vs file size at this bit rate is quite nice.

145K – 185K Variable

Exported at 145K-185K Variable bit rate with 44.1K sampling. There is little to no perceivable distortion in this audio even in the musical introduction. A larger assortment of musical instruments comprising a full band or orchestra may exhibit some minor issues, but not enough to cause serious displeasure. A direct comparison of the 128K Constant bit rate yields a technically higher quality audio file, but with little perceivable difference (if any). The file size, in comparison, dropped from 31Megs down to 26Megs for a mostly spoken audio file. The audio quality vs file size at this bit rate is acceptable for music based podcasts, but may be slightly larger than desired for spoken podcasts.

170K – 210K Variable

Exported at 170K-210K Variable bit rate with 44.1K sampling. There is little to no perceivable distortion in this audio even in the musical introduction. A larger assortment of musical instruments comprising a full band or orchestra should not exhibit any noticeable distortion, except in extreme cases. A direct comparison of the 128K Constant bit rate yields a higher quality that is imperceptibly improved with a spoken podcast. A full orchestra or other large arrangement of instruments may show some perceived improvements. The file size, in comparison, dropped from 31Megs down to 30.4Megs for a mostly spoken audio file – a negligible difference, at best. The audio quality vs file size at this bit rate may be acceptable for music based podcasts, but is too large for spoken podcasts.

220K – 260K Variable

Exported at 220K-260K Variable bit rate with 44.1K sampling. Good luck finding any perceivable audio quality problems with this one. A direct comparison of the 256K Constant bit rate yields approximately the same audio quality, though it is technically slightly less. The file size, in comparison, dropped from 62Megs down to 39.2Megs for a mostly spoken audio file – a significant reduction. The audio quality vs file size at this bit rate is not acceptable for streaming. The file is simply too large for the amount of content contained within it.

The (Somewhat Surprising) Results

As you would expect, the low-end results of the 16K bit rate are attrocious. If it wasn’t for the ridiculously small file size, this would never even be an option. In fact, I’m going to call it not an option anyways, in spite of the file size. The audio quality is simply horrible. I wouldn’t inflict that on anyone.

I’ve always known that a variable bit rate .mp3 file was going to be smaller, but I had never directly compared the file size and audio quality with a constant/variable counterpart before. The ability to keep roughly the same file size while increasing the over-all quality was a nice effect when comparing 64K Constant to 80-120K Variable, for example. The Variable is slightly larger, but with the increase of quality and negligible file size gain, this was a huge win for variable bit rates!

The other surprising result is how little difference there is, if there is any perceivable difference, between 128K and anything above it. At least with spoken podcasts I don’t see any need to ever use anything above 128K Constant rate, or the near equivalent quality of 80K-120K Variable rate.

Revised Recommendation: 96K or 128K Constant Bit Rate

I had originally written this post to recommend 80K-120K Variable Bit rate, based on the audio quality and file size. But I need to revise that based on some conversation with Ray Ortega and Rob Walch (from LibSyn) over on my Google+ post.

Rob points out several problems with audio players as recent as IOS6, and links to several posts where problems and solutions are discussed.

In the SoundManager 2 note, specifically, Rob points out the following for the SoundManager 2, HTML5 player.

“MP3 Format Caveats

Additionally, some very low and very high bitrate MP3s, and Variable Bitrate (VBR) MP3s may play either too quickly or too slowly (see “the chipmunk problem”); if you are encountering this issue, try re-encoding at a different bitrate (between 64 kbps and 192 kbps, for example.) Using Constant Bitrate (CBR) encoding may also alleviate this problem.”

Given that, I’m going to revise my recommendation:

Use 128K or 96K Constant Bit Rate unless you have a good reason not to.

On Variable Bit Rates

Even after revising my recommendation, I’m going to leave this note here, as to what my original recommendation was. At the very least, it will let you the reader see where I was coming from originally vs now.

If I were going to recommend one signal bit rate to encode your podcast .mp3 files, it would be 80K-120K Variable bit rate. The audio quality vs file size is superb, with little to not distortion or wobbles in a mostly spoken podcast.

If you press me hard enough, I’ll admit that I want the 145K-185K Variable rate for music based podcasts. The additional layers of sound found in music can present some warbles and distortion on the high end frequencies at lower rates. Honestly it’s probably not going to bother anyong at the 80K – 120K rate, but I prefer to be safe when it comes to my music encoding.

And even at this, there are podcasters out there who have switched over to variable bit rates with no issues. John Galloway for example, has been serving more than 9,000 listeners of his podcast in variable bit rate for over a year.

One Size Does Not Fit All

Clearly there is some dispute over the constant bit rate vs variable bit rate. I’m inclined to lean toward what Rob Walch from LibSyn is saying at this point. But I’m also a technologist that understands the curve of old technology falling off the market and being replaced by newer standards. In the end, you should compress your audio files according to your podcast needs, weighing file size vs audio quality and available bandwidth. Whether this means constant or variable bit rates is up to you.

Happy podcasting, all you Signalers!