Music files for big stations?

[chriscollins]

1. WAV (naturally)
2. Lossless codecs (FLAC, APE, etc)
3. MP2 at VERY high bitrates (384 kb/sec)
4. MP3

High bit .mp2 sounds better than .mp3 to me as well.

MP2 is also less sensitive to cascading with another codec that might happen down the line... As is AAC (LC). MP3 is actually the worse bit reduction codec you can use...

Our only mp3 is the commercials, which is the norm now... I do blow them back to .wav, as my Automation is happier with .wav. When a file gets blown up in my building, it gets a special tag I can search that shows MP3SOURCE.

That's an excellent practice! So many WAVs (and even FLACs) originate from poor MP3 files and people assume just because it's WAV, it's good audio quality. Unfortunately, that's no guarantee. Quite often you can tell by the poor sound and the artifacts, but people on the station just don't bother to listen or are not sensitive to that (which doesn't mean your listeners aren't!). So having a tag is a very useful technique.


Regards,
Goran Tomas

I actually came up with that idea in the early days of digital delivery. Music would 'leak' to us from the promo people as mp3. Then, when I got the CD, I would have a method to know to replace it. Now, it is normal to get .wav at the same time, but I left all my mp3 import rules the same, so I still get the tagging.

 

[satech]

Also if you must encode an MP3 yourself, set the encoder to use a lowpass filter of 15 kHz, because there is no sense in having the codec waste bits on high frequencies that will just get filtered out by your audio processor down the road anyway. (And if you look at an MP3 on a spectral graph, even at 320 kbps it can only reproduce audio up to 16 kHz accurately -- anything above that just comes out as a lot of garbage!)

 

[gtomas]

An even better idea is to encode at 32kHz sample rate...


Regards,
Goran Tomas

 

[Goat Rodeo Cowboy]

An even better idea is to encode at 32kHz sample rate...

What a beautiful piece of heresy to feed to people who have a little bit of audio-phile blood in their veins. We keep designing combinations that give us the finest sound possible, and then we get a nosebleed because high frequency artifacts are sickening. And along comes Goran with a solution that does not require learning one more exotic software regime. Just encode at 32 kHz.

Priceless!!!

 

[Kmagrill]

When an mp3 is encoded at 32kHz, what's the actual passband of the audio?

One would expect it to be nearly 16kHz, but I think it's really somewhere around 13kHz, which is slightly lower than I'd want to put on the air.

 

[gtomas]

What a beautiful piece of heresy to feed to people who have a little bit of audio-phile blood in their veins. We keep designing combinations that give us the finest sound possible, and then we get a nosebleed because high frequency artifacts are sickening. And along comes Goran with a solution that does not require learning one more exotic software regime. Just encode at 32 kHz.

Priceless!!!

If you've read my posts so far, you would have noticed that I am a big proponent of uncompressed, linear source material at CD quality. And audio quality in general.

But, if you're going to encode it to something, in this case MP3, it's no longer high quality or audiophile in any way. If you're going to do it, do it in a way that maximizes audio quality. If you have a spot for example that's made to be played on a radio, it will not be broadcast with more than 16kHz (unless you're using Omnia 9 that is in which case you can get to 17kHz)

If you have a streaming with anything less than 128 kbps btw, you would also benefit from having no audio above 16kHz (artifacts wise).

44.1kHz, 16 bit, stereo = 1411 kbps going to MP3 at 128 kbps is 11:1 reduction ratio
32kHz, 16 bit, stereo = 1024 kbps going to MP3 at 128 kbps is 8:1 reduction ratio

Get the point? Believe it or not, it actually will be a better audio quality at 32kHz sample rate! Try it. And check your 128 kpbs MP3 files at 44.1k to see how much "audiophile" HF you have above 16kHz...


Regards,
Goran Tomas

P.S. The only reason not to do it is if you have an automation system that can't play different sample rates at the same time...

 

[gtomas]

When an mp3 is encoded at 32kHz, what's the actual passband of the audio?

One would expect it to be nearly 16kHz, but I think it's really somewhere around 13kHz, which is slightly lower than I'd want to put on the air.

It's 16 kHz. I can't believe this is such a big news and people haven't played with it before...

Do me a favor - open your editing software that you use. Make a new 44.1kHz, 16 bit, stereo file. Generate a tone sweep from 10 Hz to 22 kHz. Save that to MP3@128 kbps with 44.1kHz sample rate. Now save the original tone sweep also to MP3@128 kbps but with 32kHz sample rate.

Open both MP3 files in the audio editor. See how far the audio is present in both files... Did you gain or lose something in terms of frequency response?


Regards,
Goran Tomas

 

[satech]

But when decoding the 32 kHz MP3 file, you might want to upsample the output back up to 44.1 kHz, because many PC sound cards use an anti-aliasing filter with an overly shallow slope, so it starts rolling off the audio at only 12 kHz when playing back audio at 32 kHz sampling rate (and likewise rolling off the audio at 8 kHz when playing 22.05 kHz sampling rate). At 44.1 kHz sampling rate, the rolloff point of this filter moves up to 16 kHz or so, where it has much less of an audible effect.

 

[Kmagrill]

But when decoding the 32 kHz MP3 file, you might want to upsample the output back up to 44.1 kHz, because many PC sound cards use an anti-aliasing filter with an overly shallow slope, so it starts rolling off the audio at only 12 kHz when playing back audio at 32 kHz sampling rate (and likewise rolling off the audio at 8 kHz when playing 22.05 kHz sampling rate). At 44.1 kHz sampling rate, the rolloff point of this filter moves up to 16 kHz or so, where it has much less of an audible effect.

Hmmm.. Well, If I load into Coolledit a 20Hz to 20kHz sweep and encode it at 320kbs at 32kHz with a cutoff filter of 16kHz, I get a waveform that starts rolling off at 14kHz, is quite attenuated at 15kHz and gone by 16kHz. Not bad, but the affect might be noticable. Probably, the benefit of adding bits in the main passband outweighs the minor loss of HF energy near the upper filter.

However, the anti-aliasing filter problem seems more significant. If we take an audio file with HF content to 14kHz and pass it through a soundcard with a 12kHz anti-aliasing filter, that's going to result in noticable HF attenuation. Obviously, one solution is to make sure that you're using a sound card with filters that more closely match the passband. Even a 12kHz filter for a 16kHz sample rate is a pretty amazing filter. That's dozens of dB per octave, so what sound cards have even better filters? Does anyone keep track of that particular spec?

In addition to the aliasing filter issues of the common PC sound cards, is the mp3 spectral masking as efficient at a 32kHz sample rate as it is at higher rates for dynamic content? I seem to recall reading, very long ago, that the mp3 codec can have problems with unexpected attenuation of frequencies above 13kHz when a 32kHz sample rate is used. Perhaps that's not really true, or maybe it was true of the encoders of 10 years ago, but not now or maybe the author was only referring to the soundcard anti-aliasing filter issue. Does anyone authoritatively know if that's the case or not?

 

[Dale H. Cook]

what sound cards have even better filters?

I would expect pretty much any professional sound card. As a rule input-to-output on a pro card with 44.1 kHz sampling will measure essentially flat to 20 kHz, and with 48 kHz sampling essentially flat to 24 kHz.

Does anyone keep track of that particular spec?

After looking through my collection of manuals for Antex, AudioScience and Digigram pro audio cards I saw no specs for anti-aliasing filters. In any case the overall performance of an audio card can easily be measured with the appropriate equipment (say, an Audio Precision system).

Dale H. Cook, Market Chief Engineer, Centennial Broadcasting, Roanoke/Lynchburg, VA
http://plymouthcolony.net/starcityeng/index.html

 

[satech]

I think once you factor in the downsampling/upsampling and anti-alias filter rolloff, a 32 kHz MP3 file at reasonable bitrates (128+ kbps) is going to come out worse than, or at least no better than, a 44.1 kHz MP3 file of equal bitrate and audio bandwidth (up to 15 kHz).

Also I recall that the lower the sampling rate, the worse MP3's "pre-echo" artifacts become. This was clearly evident with the "MP3pro" format, which claimed that a 22.05 kHz, 64 kbps MP3 file with SBR (fake treble) was supposed to sound the same as a regular 44.1 kHz 128 kbps MP3 file. But even if you could ignore the synthetic SBR treble, MP3pro's transient response to things like snare drum hits sucked, because of the bad pre-echo problems of using the low 22.05 kHz sampling rate with MP3.

 

[gtomas]

Guys, I don't know what kind of soundcards you are using, but a proper soundcard has a response of 16kHz with 32 kHz sample rate. Not 12k, not 14k, but 16kHz.

The soundcard also doesn't have to switch the sample rate it operates on necessarily, but can natively run at 44.1kHz or 48kHz and play 32kHz audio. It's called sample rate conversion and it can be done automatically (as an example).

I don't know what kind of MP3 encoders or editors you are using either. This is a picture of a tone sweep in Adobe Audition saved as MP3@128kbps with the encoder set to 32kHz sample rate (green line) and then saved as MP3@128kbps with the MP3 encoder set to 44.1KHz sample rate (red line). Spot a big difference?

If you're not comfortable with 32kHz sample rate, fine. But let's not come up with war stories about 32kHz sample rate and codecs running at that sample rate...


Regards,
Goran Tomas

 

[chriscollins]

Guys, I don't know what kind of soundcards you are using, but a proper soundcard has a response of 16kHz with 32 kHz sample rate. Not 12k, not 14k, but 16kHz.

The soundcard also doesn't have to switch the sample rate it operates on necessarily, but can natively run at 44.1kHz or 48kHz and play 32kHz audio. It's called sample rate conversion and it can be done automatically (as an example).

I don't know what kind of MP3 encoders or editors you are using either. This is a picture of a tone sweep in Adobe Audition saved as MP3@128kbps with the encoder set to 32kHz sample rate (green line) and then saved as MP3@128kbps with the MP3 encoder set to 44.1KHz sample rate (red line). Spot a big difference?

If you're not comfortable with 32kHz sample rate, fine. But let's not come up with war stories about 32kHz sample rate and codecs running at that sample rate...


Regards,
Goran Tomas

You beat me to it... My Digigram cards work all the way 16KHz when I use a 32 KHz file... Audiovault uses a software codec for SRC. No problems there.

 

[Kent T]

MP3 should not be used for any broadcast purpose but voice elements. It does not cope well with multi generation conversions in a modern digital air chain. MP2 is a better alternative or AAC better yet if compressed audio files must be used.

 

[satech]

Guys, I don't know what kind of soundcards you are using, but a proper soundcard has a response of 16kHz with 32 kHz sample rate. Not 12k, not 14k, but 16kHz.

Yes, it will have some response up to 16 kHz, but not flat response. No anti-aliasing filter (even digital) is perfect in terms of both frequency and phase response. Isn't this the whole reason digital audio processors got away from using 32 kHz sampling rate?

I don't know what kind of MP3 encoders or editors you are using either.

I've tested MP3 for many years in many different scenarios, and there are really very few situations in which using a 32 kHz sampling rate makes any major benefit compared to using 44.1 kHz at the same bit rate. At 96 kbps it's a wash either way, but at anything higher than that, 44.1 kHz has the clear advantage, while at anything less than 96 kbps, I've always gotten best and most consistent results by stepping down all the way to 22.05 kHz (although for 44.1 kHz mono, 80 kbps usually sounds quite acceptable).

 

[Goat Rodeo Cowboy]

Yes, it will have some response up to 16 kHz, but not flat response.

I am in the process of getting acquainted with Adobe Audition CS6. Goran had a link in his post to a screen-shot showing the Frequency Analysis of the sweep-tone. FLAT!

So I cranked up my CS6, found the menu to generate the sweep-tone, changed to file to a 32 kHz sampling rate and played with it. Then opened up the Frequency Analysis to see if the upper end dropped off. FLAT! But does it have artifacts? Someone else will have to make that judgement. I don't think my ears function well enough up in the 15 to 20 kHz area to know.

 

[satech]

So I cranked up my CS6, found the menu to generate the sweep-tone, changed to file to a 32 kHz sampling rate and played with it. Then opened up the Frequency Analysis to see if the upper end dropped off. FLAT!

That's just software-based digital processing, so of course it's going to be (and should be!) as close to perfect as possible. You have to measure your sound card's external analog loopback -- out through the its DAC and then back in through its ADC -- to get an accurate measurement of what your ears are actually hearing at various sampling rates.

 

[Goat Rodeo Cowboy]

That's just software-based digital processing, so of course it's going to be (and should be!) as close to perfect as possible. You have to measure your sound card's external analog loopback -- out through the its DAC and then back in through its ADC -- to get an accurate measurement of what your ears are actually hearing at various sampling rates.

Yes, and sometime tonight or this week, I am going to do that. I am going to create a file with some tone pulses, or break up the sweep file into short bursts and find the best set of earphones in the house and test my hearing a bit, just for fun.

But the main thrust of this discussion has been the comparison of various protocols of sample rate and bit depth of mp3 files vs. "the gold standars" a.k.a. .wav files. I guess it is possible to to take a a digital audio file and pass it on to digital audio devices and pass it on to a transmitter that accepts digital in and never deal with the soundcard DAC and ADC process. We let the transmitter deal with that process, or if the station is digital, we let the listeners received handle that task.

The basic question we have been kicking around the room like a soccer ball is the narrow question of "how badly does mp3 affect my audio content, and can that be minimized by the choice of sample rate, bit rate and bit depth?" Once the person at any radio station (or streaming operation) decides to store the entire music library in one consistent protocol, we still have that DAC process in the sound card to analyze which is another topic.

Maybe the reason some people are so picky while some other people are so flexible is that their soundcard has more effect on the output quality (including artifacts) than does the files storage technology. The station with a $1,500 sound card may play .mp3 files that sound better than the station down the street playing .wav files through the built-in card on a consumer PC purchased at a big-box store.

 

[Kmagrill]

So I cranked up my CS6, found the menu to generate the sweep-tone, changed to file to a 32 kHz sampling rate and played with it. Then opened up the Frequency Analysis to see if the upper end dropped off. FLAT! But does it have artifacts? Someone else will have to make that judgement. I don't think my ears function well enough up in the 15 to 20 kHz area to know.

Another test might be to take a .WAV 20Hz to 20kHz sweep and encode that directly to an mp3 with a 32kHz sample so that it's all done in one step. I cannot speak to Audition, but CoolEdit Pro's mp3 encoder produces a nice flat output to 14kHz that tapers off rapidly from 14kHz to 16kHz where the output is down more than 70dB.

 

[satech]

I discovered that the "SoundMAX Integrated Digital HD Audio" built into my Lenovo ThinkCentre desktop has something even worse than a bad anti-aliasing filter when playing 32 kHz audio -- it has no anti-aliasing filter at all!

This was the result of saving a 0 to 16 kHz sine wave tone sweep to a 32 kHz, 16-bit stereo WAV file, and playing it back while simultaneously recording it at 44.1 kHz via the sound card's "Stereo Mix" input in Adobe Audition:

http://i48.tinypic.com/20qnek2.png

And this was the result of saving a 0 to 16 kHz sine wave tone sweep as a 44.1 kHz WAV file, and recording it at 44.1 kHz:

http://i46.tinypic.com/jr6y2s.png

There you can see the anti-aliasing filter does exist and works properly at 44.1 kHz.

It also showed a complete lack of anti-aliasing at a sampling rate of 16 kHz, although at 22.05 kHz it clearly does use a pretty good filter -- not as steep as an NRSC brickwall, but still acceptable:

http://i50.tinypic.com/2hxu49k.png