.mp3 Compression

[Brilliant_Marconi]

For posting mono .mp3 audio to web-sites (archive audio library) - what are the recommended compression settings for mono audio files without noticeable audio quality degradation? Obviously, we want to compress the audio files are much as possible for ease in downloading by our users, while maintaining decent audio quality. Yes, a fine balance. Suggestions would be appreciated. Thanks very much.

 

[chriscollins]

A 128k Mono would sound virtually as good as a 256k stereo file.

If it were me and it were mono, 128 would be my choice.

 

[RadioMan203]

I say it really depends. I have quite a few airchecks on a radio station tribute site I run and with limited storage space I needed to draw a compromise between storage space and acceptable fidelity.

With my mono airchecks I've gone between 32kbps and 48 kbps. With stereo airchecks I've gone between 48 and 64. Granted, the main focus of these audio files are the voice portions which is why I could skimp on the bitrate more.

If you're archiving musical performances or want better quality sound with less aliasing I would go no lower than 128 although I've heard 96k sound acceptable.

Something else to keep in mind is if you're on the hook for audio bandwidth as part of your site's total bandwidth. Depending on the amount of traffic your site draws, If your files are hosted elsewhere that may be less of a concern.

 

[Goat Rodeo Cowboy]

Marconi: you didn't tell us anything about your program content. Music? Speech?

And you didn't tell us anything about your audience profile: audio enthusiasts? People listening on serious audio systems or people listening on $24.95 MP3 players strapped to their arm while jogging.

I edit and produce mp3 recordings of sermons for a church. I have been using 40 kbps (mono) for the spoken word... no music. I was in hopes that would work o.k. for someone listening on-line via a dial-up connection.

I did one for a regional organization this past week and bumped it on up to 64 kbps just for grins. I may do that for the regular series and sit back to wait for any blowback from anyone who has a slow connection.

I would make recordings at 40, 64 and 96 and listen to them on devices comparable to what your audience is likely to use and see for yourself how it works.

(I tried some of these recordings several years ago at 20 kbps and 24 kbps. I found both of those to be just a bit flawed to my taste. I moved on to 32 and 40 and found both of them acceptable. Your mileage may vary.)

 

[Brilliant_Marconi]

All good points....audio is mostly spoken word....the majority are sports play-by-play audio files. Most listening will be done via the computer or downloaded and saved to a hard drive or burned to CD for personal archiving (mom/dad getting a copy of son Timmy throwing a touchdown pass).

 

[RolfTaylor]

Compression ratio is also determined by the sample rate. For voice you would never want to use a sample rate >32 kHz. Often lowering it to 16 kHz will make a big improvement in quality if the bit rate is 64 kbps or lower.

128 kbps is probably the best overall "safe" for mono IMHO, but it depends on if you will be streaming versus file downloads. For streaming you'll want to get down below 64 kbps which is fine if you back down on the sample rate.

That said, voice is harder to encode than most music (though MP3 handles glockenspiel music particularly poorly)

Some voices (for example Susan Vega) are much harder than others. Mine falls in this category, which we found while developing the Zephyr Xport!

cheers

R8)LF

 

[Tom Wells]

It also depends on what you personally find acceptable.
I used Pod-O-Matic until they downgraded my mono 128k files to 96k in playback.
Not worth it anymore.
I now hear my own 128k mp3s on the air, and some sound OK, but others have beginnings of the swirl effect,
so the minimum I use now dubbing vinyl is 192k. This is on an AM, and the difference IS audible
on the car radios.

We accept a LOT more distortion in the presentation of music because we listen to so much recorded music.
But we hear live voices everyday and when something is wrong with the sound of a voice, our
brain is very quick to tell us that something is wrong with our perception or the presentation.

 

[Kmagrill]

Compression ratio is also determined by the sample rate. For voice you would never want to use a sample rate >32 kHz. Often lowering it to 16 kHz will make a big improvement in quality if the bit rate is 64 kbps or lower.

128 kbps is probably the best overall "safe" for mono IMHO, but it depends on if you will be streaming versus file downloads. For streaming you'll want to get down below 64 kbps which is fine if you back down on the sample rate.

That said, voice is harder to encode than most music (though MP3 handles glockenspiel music particularly poorly)

Some voices (for example Susan Vega) are much harder than others. Mine falls in this category, which we found while developing the Zephyr Xport!

cheers

R8)LF

There is a relationship between files size, sample rate and compression ratio, but to clarify a bit more, there's not a direct relationship between compression ratio and sample rate. Rather, compression ratio and sample rate are both factors in file size (and stream bandwidth, too). Each also have their own effect on percieved quality. There is also the fact that most stereo files are joint-stereo. While a true stereo file requires 2x the size of a mono one, a joint-stereo file is primarily mono with only the discrete stereo information encoded individually on each channel. The result is a file that performs like it is encoded at a higher rate while preserving most of the stereo image.

For music, 32k sample rates produce FM quality audio. There's really no need for 44.1 or 48kHz sampling and your low bitrate mp3 is going to discard almost everything ultrasonic anyway. So, given that you can probably sample at 32kHz or even 24kHz, a mono file should perform very well at 80kbs (or even 64kbs for a 24kHz sample). You might even be able to get down to 48k.

 

[GeneSavage]

As others have said, it really depends on your source and your target.

For posting on-demand voice, 40 kbps mono should be perfectly acceptable. The only time that might not be the case is if you have a noisy source (a cassette tape, for example).

For posting on-demand music, I recommend bumping up to 48 kbps mono. The longer tones make compression more obvious, so you really need to bump it up.

For stereo, I'd run no LESS than 96 kbps, and be forewarned that slow ballads will gurgle and growl at this bitrate... most casual listeners don't seem to notice.

For streaming, 128 kbps stereo seems to be the "industry standard" although you will still occasionally hear swirling and swishing.

For demanding stereo content and picky listeners, 192 kbps is an easy choice to make for extremely high fidelity. So claim to hear the difference between 192 kbps and 256 kbps, but I've never met anyone who could prove it to me.

Two things to remember: #1, LAME is going to make a better sounding file at the same bitrate than the Fraunhofer-Gesellschaft (FhG) encoder built into Adobe Audition / Cool Edit Pro. It really IS worth it to save as a .wav file to your hard drive, then encoder with the LAME codec. I use RazorLame as the front-end for LAME.

#2, the higher the bitrate, the higher the fidelity / the longer the download / the most likely the live stream will stutter. AAC+ is coming on strong, but not all players support it. If enough of your listeners supported AAC+, it will give MUCH higher fidelity at MUCH lower bitrates.

Good luck!!!

 

[gtomas]

There is a relationship between files size, sample rate and compression ratio, but to clarify a bit more, there's not a direct relationship between compression ratio and sample rate.

No direct relationship between compression ratio and sample rate?

44.1 kHz, 16 bit, stereo = 1411 kbps compressed to 128 kbps, gives compression ratio 11:1.

32 kHz, 16 bit, stereo = 1024 kbps compressed to 128 kbps, gives compression ratio 8:1.

Rolf is absolutely correct that lowering the sample rate reduces the "compression" (reduction of data) for a given target bitrate, simply because it reduces the size of the data to begin with!


Regards,
Goran Tomas

 

[gtomas]

AAC+ is coming on strong, but not all players support it. If enough of your listeners supported AAC+, it will give MUCH higher fidelity at MUCH lower bitrates.

HE-AAC (aka aacPlus or AAC+) will not give you higher fidelity per se. At very low bitrates it will give you higher audio bandwidth compared to other codecs, but the high band of HE-AAC is always generated and therefore artificial.

AAC will give you higher fidelity compared to MP3, as it is better and newer codec and 30% more efficient than MP3. AAC was found to be transparent (indistinguishable from the source) at 128 kbps stereo, on EBU listening tests.

As people seem to confuse the two, AAC and HE-AAC (AAC+) are not the same codecs. HE-AAC is an extension of AAC designed for low bitrate streaming. As such, it performs better at very low bitrates (at or below 64 kbps) compared to AAC, but performs poorly at higher bitrates (at or above 96 kbps) when compared to AAC.

In this day and age of ADSL, when minimum download speeds are 2 Mbps and typically 4-8 Mpbs, there is really no need to sacrifice audio quality for bandwidth. If you want to stick to MP3, I would go with 128 kbps or minimum 96 kbps for mono.


Regards,
Goran Tomas

 

[Kmagrill]

Rolf is absolutely correct that lowering the sample rate reduces the "compression" (reduction of data) for a given target bitrate, simply because it reduces the size of the data to begin with!

I was just trying to suggest that there's no specific formula for compression vs sample rate. In other words, I read the earlier post as stating that the sample rate determines the compression ratio, which I would not characterize in that way because it is possible to use high bitrates for low sample settings with no affect. Sometimes, it is even possible to use low bitrate coding for fairly high sample rates without being too affected. It is definately true that there is an important indirect link between the two, however. Probably, I was just tired, because I re-reading it, I see that it can be read to suggest that the sample rate affects the choice of compression rates. That is to say by starting with a smaller sample, you can elect to use a more aggressive codec setting because it has less spectrum to contend with. Thus, a lower quality setting will often produce just as good an outcome as a high setting because the bits are spread out over a smaller area. So, to the extent that reducing the sample (generally) allows one to apply a more aggressive compression ratio, you are correct.

 

[Goat Rodeo Cowboy]

Goran: from time to time you pay homage to the superior features of AAC and HE-AAC. I know Bob Orban also writes occasionally about AAC+ and on my old computer I downloaded a software he made available. Here is my observation... or if you please: here is my COMPLAINT.

There is too little information available about how to use the formats, and how to troubleshoot when conflicts arise. My old computer crashed and burned about a month ago and I am for a brief period enjoying the joys of a new, clean pristine machine. The intake of streaming is a wild no-mans land!!! I know the minute I put AAC+ on my new machine (if I do) that immediately I will find that some sites that put plain old mp3 up will cause my machine to come to a halt, throw up a big info box about "Can not find software to play AAC+".

That's really cool, machine. I have AAC+ software on my machine and I routinely go to certain sites that feed AAC+ and it comes in fine. But machine, right now I am trying to stream old fashioned, plain vanilla mp3 and you have your panties in a wad and refuse to stream the mp3 because you have this unfounded inferiority complex about not being able to digest AAC+, which you actually do on occasion.!

Goran: my skills levels in these area are somewhere on the dividing line cusp between serious amateur and journeyman professional. I have beat the Internet to death trying to figure out why I can click on this radio station's live stream and have success, and when I click on the next station's live stream, I get the ugly message box described above. What sources am I failing to find? What book do I need to buy?

 

[gtomas]

Well, AAC would be my codec of choice as far as best audio quality is concerned.

But I don't really understand what do you mean when you say "when I put AAC+ on my machine"?

May I suggest you download WinAmp, a great little player that will play MP3 streams and AAC streams and HE-AAC streams without a hitch. Another good one is VLC Media Player.


Regards,
Goran Tomas

 

[gtomas]

Or I might be missing a joke here...


Regards,
Goran Tomas

 

[RolfTaylor]

Ksmagrill, I think you are still confused. Here is how informed decision-making on this process works:

Determine required audio bandwidth (you may wish to re-evaluate this later in the process). Remember, the spectrum above 15 kHz is not going to be heard by older listeners, nor those that do not have high-end speakers.

Select a sample rate

Determine required resolution - odds are you will want 16 bits. You will NOT want to use anything greater than 24 bits since that is simply shooting yourself in the foot.

Determine if mono or stereo is required

Determine the target bit rate.

Calculate the compression ratio

Choose a codec - Here are some typical codec choices for various ratios:

2:1 - Huffman encoding, etc
4:1 - APT-X or MPEG Layer 2, AAC
6:1 - 8:1 MPEG Layer 2, AAC (do not use Layer 3 at ratios less than 8:1)
8:1 - 14:1 MPEG Layer 3, AAC
12:1 to ~18:1 AAC, Layer 3 plus
>/= 18:1 HE-AAC
>/= 30:1 voice-only codecs

If no codec is available for the ratio required, you need to go back and reconsider other factors, or accept artifacts

Of course there are other factors that come into play, but the above is the way to make an informed decision.

TIP - to much bandwidth is frequently an area where poor decisions are made. In many cases the addition of high end will simply push the ratio to a point that is too high for the available codec.

TIP2 - As a rule of thumb, any ratio greater than 7:1 is likely to have issues with cascading. Good practice demands linear source material and eliminating multiple coding passes. (APT-X is a big exception to that rule, but since it only operates at 4:1 the rule still stands)

I hope this is helpful.

Rolf Taylor
Twins Ace Hardware

Rolf is absolutely correct that lowering the sample rate reduces the "compression" (reduction of data) for a given target bitrate, simply because it reduces the size of the data to begin with!

I was just trying to suggest that there's no specific formula for compression vs sample rate. In other words, I read the earlier post as stating that the sample rate determines the compression ratio, which I would not characterize in that way because it is possible to use high bitrates for low sample settings with no affect. <snip>

 

[Kmagrill]

All good points Rolf. I'm not confused, but maybe I'm not expressing what I'm saying very well. I believe that your process is pretty much what I always do. I simply point out that you can choose to use less compression (or more depending upon your tolerance for grunge) at any given sample rate. Many would consider using less compression to be a waste of bandwidth or disk space, but it is a subjective matter and because the results are subjective, different formulas can be applied by different people to achieve the result that they desire. Thus, I contend it's 75% science and 25% art and therefore there is an indirect link between between sample rate and compression ratio, though I agree that sample rate can have have a bearing on the choice of compression when one is trying to get maximum efficiency.

 

[littlejohn]

>Rolf Taylor
>Twins Ace Hardware

Uh, Rolf, are you now the Duplicate Key Jockey?

 

[satech]

The funny thing is, back in the '90s, RealAudio used to be perfectly acceptable for mono voice content at very low bitrates (16-20 kbps), and to my ears, still is... but since it is a proprietary format and long out of date, nobody wants to (or should!) support it anymore. Still, it's strange that we had a perfectly good solution for streaming Internet audio 15 years ago, which could deliver acceptable quality on as low as a 28.8 kbps dialup connection... and now, we've stepped back to an even older and less efficient format (MP3), just because it's "open source" and portable media players have put it into vogue. Such is progress...

 

[chriscollins]

The funny thing is, back in the '90s, RealAudio used to be perfectly acceptable for mono voice content at very low bitrates (16-20 kbps), and to my ears, still is... but since it is a proprietary format and long out of date, nobody wants to (or should!) support it anymore. Still, it's strange that we had a perfectly good solution for streaming Internet audio 15 years ago, which could deliver acceptable quality on as low as a 28.8 kbps dialup connection... and now, we've stepped back to an even older and less efficient format (MP3), just because it's "open source" and portable media players have put it into vogue. Such is progress...

You can blame Real in a sense for some of that. The death of them as an encoding format was the bloated player that got bigger and bigger and the built in wma streaming within windows. Having a wma server is an easy setup.

On the mp3 front, it's what caught on, so we have to adapt. You want the most people to hear your product and who can't play an mp3?