Technical Quality of digital broadcasts :(

[Dave]

Health Warning

Unless you are an audio "techie", this post may damage your health - do not read it!!!

Tonight I've spent a lot of time recording and digitising Hayley's performance on BBC Radio 2's "Friday Night is Music Night". This was a live broadcast from London and I was able to record it in two ways - from Satellite Digital radio and from a high quality FM tuner directly onto my PC. Although the FM recording was analogue, all my PCs have Soundblaster Live soundcards which digitise at "hi-fi" quality.

The Satellite digital radio recording was initially made losslessly within the satellite receiver, then copied to my PC using the same analogue link as the FM broadcast. All subsequent processing and comparisons were done in lossless .wav format (apart from the final conversion to a 192 kbps .wma file for uploading to the Internet).

Therefore, I had a unique opportunity to compare recordings of the same "live" programme material obtained from a) an analogue FM radio source and 2) a good quality digital radio source.

Well, I eventually completed a second version of the recording (from digital radio) and set about comparing it with my original recording from FM radio, using A/B switching on my PC - and waveform analysis/measurements too. There follows a summary of what I found.

Compared were:

1)Uncompressed .wav files from FM radio.
2) Compressed .wma files at 192 kbps, originally from FM radio.
3) Compressed digital radio - ostensibly at 192 kbps (but I have my doubts) and digitised as a .wav file.
4) Compressed .mp3 files posted by Gerrit from a 40 kbps Internet feed.

How loud is loud?

First, let me explain to any uninitiated readers that the loudest possible part of a digital music file is measured as 0 (zero) decibels (dB). As the music gets quieter, the volume goes negative (-2 dB is still very loud) and for practicalpurposes, the silent bits in live concerts are usually around -40 dB except on CD, where it can be much lower.


Peak sound levels (the loudest bits)

All four files were set (normalized) to peak at -1.8 decibels.

Noise floor (the "silent" bits)

All four files had a similar level of "silence" (well as silent as you ever get in a "live" concert) at about -40dB.

The quietest music

The first two notes of musical intro at the start of "Lascia" - about 2.5 seconds.

FM uncompressed -29.1dB*
FM at 192 kbps .wma -29.1 dB*
Digital Broadcast Radio -23.3 dB*
Digital Internet Radio -19.1dB*

Now, this is a very gentle, quiet intro but the digital broadcasts have already boosted its audio level by 5.8 to 10.0 dB! That is a lot and you can see how digital broadcast radio distorts the reality of very quiet music - lowers the dynamic range.

Slightly louder music but still fairly quiet

The third to the fifth notes of musical intro at the start of "Lascia" - about 3 seconds: just before Hayley starts singing:

FM uncompressed -20.6 dB*
FM at 192 kbps .wma -20.6 dB*
Digital Broadcast Radio -14.7 dB*
Digital Internet Radio -11.5 dB*

This bit was played slightly louder by the orchestra but as you can see, the digital broadcasts whacked the volume up by 5.9 dB to 9.1 dB compared to FM radio.

I think both the above comparisons at the start of Lascia show that the gentleness of this intro is completely lost by digital and internet radio.

Hayley's first word "Lascia"

Next, Hayley starts to sing - "Lascia" is her first word:

FM uncompressed -12.6 dB*
FM at 192 kbps .wma -12.6 dB*
Digital Broadcast Radio -5.1dB*
Digital Internet Radio -4.0 dB*

Once again, a big difference of 8.6 dB between the different versions. Both digital radio versions are awful - close to full volume but, Hayley is nowhere near full volume; she sings 10 decibels louder, later in the song. You cannot tell this from the digital radio broadcasts.

So what's happening here?

Well, digital radio is grossly compressing and in my opinion ruining the dynamic range that should be heard in a live classical concert. This was a BBC broadcast but other digital stations are far worse, trust me. What they do is squash the noise floor downwards to make it sound nice and quiet and with no background hiss or funny noises, but starting with the quietest notes, they ramp up the volume considerably, making quiet notes too loud. But when they get to the louder notes, they have nowhere to go so they have to make it all pretty well the same loudness.

This is serious audio compression and peak limiting and frankly, to anyone who cares about hearing good music exactly as it was intended, it sucks.

But by making everything sound louder, it is initially impressive. Just like certain "loud" crossover singers I could name. But I'm sorry, it's not for me, not unless I only want it as background music or in the car.

And as far as digital radio is concerned - yes it's easier than FM radio to pick up with no interference - but look what it does to the fidelity. But they do not have to do this! Look at the 192 kbps windows media file I produced for Friday night is Music Night. The dynamic range is *identical* to the original uncompressed recording from FM radio. Why oh why do they do this then? For Classical music, digital radio is a backwards step in quality and it is nowhere near the quality of CDs played through a good hi-fi system because of squashed dynamic range. But, I have to admit, many people will care not one jot about any of this. Oh well.

Grrr!!!! Rant over.

Dave

PS: I may move this somewhere more suitable if it's the wrong place. I'm going for the shortest thread on the forum so I wanted it out of the way!

[postscript]

Hi Dave!

I'm not sure how many members will respond in the way I am going to. I suspect many will jaw-drop gawp and pass rapidly on... but i could be the one out of kilter!

As far as I am concerned, I've copied and pasted your post so I have it to hand, just in case you move it and i lose it.

I found your post informative (if periodically above my head, hence the need to study offline) and I admire where you are in terms of equipment and resource. Its where I once was in older technology in my student days when I was handling sound effects etc for stage work and my own personal recording library, gained from the Long Playing Record Library. I think that is what it was called. Do you remember those days, when LPs would turn up in the post bonded between two thin squares of plywood?

Anyway, thanks for sharing BUT and I suppose there is a technical problem regarding copyright [?reference Mods], anything going to be up on the board?

Peter

[alien]

More Health Warning

Very interesting analysis Dave! After looking at your post, I couldn't resist not playing around with the numbers

dave said:

Peak sound levels (the loudest bits)

All four files were set (normalized) to peak at -1.8 decibels.

Since audio compression throws away lots of information from the source audio, normalizing the audio with peak sample may not always result in the same "average" loudness after compression. Wonder what the audio would be like with RMS normalization instead of peak sample

After inputting the data you carefully collected into Excel for graphing:


The relative differences in loudness (dB) between different passages seem to be the same for your FM radio recording and Gerrit's Internet radio one. As for the wma encoded version of the FM radio recording, the loudness difference between "A" and "C" is similar to other sources. If we take a look at the difference between the loudest and quietest samples:
FM Radio: 14.6dB
FM Radio (wma): 16.4dB
Digital Radio: 18.2dB
Internet Radio: 15.1dB

It appears that with the exception of the digital radio, the difference hovers around 15dB +- 1.5dB. Could the higher loudness difference in the digital radio implies lower dynamic compression?

I haven't heard the different versions of the recording so my "analysis" is based on the numbers only and is probably flawed.

When it comes to CD recordings, I do notice newer CDs tend to sound a lot louder and the dynamic compression often result in clippings. I was checking out the Il Divo album at the store the other day and noticed there were very noticible clippings in several tracks. The same clipping problem can be found on the song's video posted on their website so clearly it's a mastering problem

Hopefully this compression trend doesn't continue. They're ruining the music! Perhaps that's their evil plan to force us to buy SACD or DVDA since they tend to be less compressed dynamically

Sorry about the rant

Allen

[Dave]

Hi guys, thanks for the comments, i didn't expect any!

I wasn't happy with the results i got for the windows media file (from FM radio) that I uploaded, so I did all the tests again and re-checked them. I found that i did my wma file a severe injustice as its dynamics are in fact identical to my original uncompressed .wav file. I have amended my original figures (in blue) accordingly. So the dynamic range that you hear when playing my file is essentially what I heard when listening to the original FM broadcast.

Therefore, the problem of dynamic compression in digital broadcast radio is one introduced by the digital broadcasters. I can understand this being done for a low bitrate internet broadcast but there is absolutely no excuse for it when broadcasting at 192 kbps, or even 128 kbps. It is clearly a policy decision by broadcasters, for what reason i know not. For the superficial impression of "loudness"? Who knows? But it most certainly is NOT hi-fidelity as broadcast. this smacks of appealing to the masses (pop music fans) but when a classical concert is broadcast as on Friday, classical music fans lose out badly.

The most significant figures from what i quoted are the final set as they show that when Hayley's singing gradually increased in volume by 10.6 dB (from -12.6 dB to -2 dB, near the peak level) in "Lascia", the digital broadcasts were able to increase it only by 3 and 2 dB respectively (from -5 and -4 dB to -2 dB) because they were already pretty well maxed out. That is a criminal level of compression on a live "classical" concert. What is the point of Hayley singing with feeling and skill - introducing dynamic subtleties as the songs were intended - if 10.6 db of dynamiv range is squashed into a barely discernable 3 or 2 dB? What IS the point?

Grrr!!!!

I will later do some tests on the classical stations...

Cheers, Dave
PS Allen, i like your rant!