Friday, 29 March 2013

BAFTA Craft Nomination 2013 - Sound Fiction

I'm still in complete shock about this, but I've been nominated for a BAFTA Craft Award in the Sound - Fiction Category for my work last year on the Accused: Mo's Story.

Well done to everyone who has been nominated - especially Michael Harrows and Gavin Finney, who I worked with on Mr. Selfridge last year, as they are thoroughly lovely people - as well as being very good at their jobs!

Well...the awards ceremony is on the 28th April, 2013 - better get a suit then!


29th March 2013

Monday, 11 March 2013

One Meter to rule them all...

Ever since owning my Teac 3340, I can always remember being fascinated with the VU metering compared to how loud something sounded to my ears; and even on the Teac 442 the section on metering in the manual for the early Tascam cassette 4 track was nearly as well thumbed as the section on bouncing down tracks. Digital meters behave so differently to the analogue ones that I grew up with, and it's somewhat reassuring that there are so many emulations of analogue meters for DAW's, even though the metering requirements for digital tracking can be so different. But then, it comes back to that quintessential question - do meters show you how loud something is?

How Loud is Now?

I suppose the first question to ask is how do we hear loudness - and the answer to that is (as far as I can tell at least) is not a simple one.

Robinson Dadson Loudness Contours.
Firstly, the ear is a non-linear device, and the frequency response of the ear is dependent on the volume at which we are listening; still the most common referenced work is by Fletcher-Munson, where as the work carried out by Robinson-Dadson in 1956 was carried out in a free field (i.e. without headphones) and has formed the basis of the most general standards - but in 2003 Suzuki and Takeshima have carried out further studies to propose a new set of loudness curves.

And that's just how loudness affects our perception of frequency - and it does show why we should set ourselves a standard, and repeatable, monitoring volume in our control rooms and headphones; so that we have the same frequency response whilst working (not withstanding ear fatigue). As for amplitude alone, our ears work in such complicated ways that we are sensitive to a massive dynamic range, and our ears can be equally damaged by a high volume transient, as long term exposure to a marginally high volume.

Analogue Kid

Not wishing to harp on about the "(good?) old days", but analogue metering was there to indicate the voltage level of the signal, with the knowledge that any rapid peaks within the system could be handled by the headroom / saturation characteristics of the equipment. The physical weight of the VU meter needle meant that it had a slow response time and that a very rapid spike would not be correctly measured on the meter - somewhat safe in the knowledge that the headroom of the electronics and saturation of tape could accommodate the spike - and even add character (dependent on the quality of the equipment); that was why so much was recorded in "the red". Also, this slow response of the needle essentially gave it an "integration" (averaging) quality, with a rise time of around 300ms, which is somewhat analogous how our ears hear things. One of the biggest problems of the VU meter was the lack of specifications, meaning that there was a massive variability across budget equipment.


The PPM (peak programme meter) was (and still is) the professional (i.e. expensive) audio meter and has a different scale and response characterises compared to the VU meter: and a wide range of scales / ballistic options. In the UK, the BBC PPM meter was the format used for broadcast sound, and upon a first look had a very different scale  (1 - 7 with no units, but with a regular 4dB spacing between the gradations) and very specific ballistics.  But again, these meters would let through very rapid transient peaks (based on the assumption that the human ear could not detect very short duration distortion) - with PPM6 (+8dBu) being the "maximum" reading.

Digital World

So, how do a VU meter and a BBC PPM sit in a digital system and on a digital scale? For a VU - the zero point (marked as 0 VU (volume units)) was defined as +4dBu or 1.228 V RMS over a 600 Ohm load; for a PPM (BBC), PPM4 is defined as 0dBu. And thanks to Hugh Robjohns and the Sound on Sound Team, there is a great visual table that helpfully ties the various scales together with the dBFS scale of the digital domain - and in fact the whole SOS article from which this image is taken is a great read.

But what is clear now is that it's not just a matter of different scales and different units, but the ballistics of the meter to suit the job - as a VU meter or PPM will not respond quickly enough to respond to the needs of the digital system, as in the digital domain there is no head room; beyond 0dBFS there is only horrible digital distortion. What's worse is that even a sample peak programme meter (SPPM) - i.e. one that measures the amplitude of each sample, and is commonly found in all DAWs - is not sufficiently accurate. There may be signal peaks between between the base rate sample points that could be 3dB greater than the measured value. This image shows how a SPPM can under read the true peak signal levels - and this comes from an excellent article by Simon Pegg from Broadcast Engineering.

In fact, it was the testing of my meters by using a 12000.1 Hz tone (at a base sample rate of 48kHz) that sparked this blog entry off - and if you want to see how SPPM's can 'mis-read' a signal, carry out this test yourself. The above soundcloud file is three constant amplitude sine waves, one at 12000 Hz, one at 12000.1 Hz and finally one at 12000.5 Hz. Drop it into your DAW or edit suit running at 48kHz, and you'll see your meters displaying a varying amplitude even though the amplitude of the signal is constant. It's because the SPPMs are running at the base sample rate and 'missing' the peaks in the wave form - effectively seeing beating, as the sample rate and the test tone are a fraction of a function of each other. This video from eyeheight explains it really well.

True Peak and EBU R128

This under reading from SPPM means that when mastering up to 0dBFS, that these inter-sample peaks will be being clipped, and that for broadcast the actual transients will be beyond the specifications as identified by the broadcaster. So the ultimate digital meter is the True Peak meter (measured as dBTP) which over samples the digital signal to accurately measure the true peaks within the signal (these meters should adhere to ITU - R - BS1770, and should have 4x over sampling as a minimum). But - somewhat pragmatically - these inter sample peaks are so short, so can we really hear them? And perhaps a simpler solution is just to allow more headroom within our digital mixing.

And now for broadcast, we have a new loudness specification that (hopefully) all broadcasters will be adhering too - EBU R128, (Florian Camerer does as great job of explaining it all here), that introduces the concept of loudness normalisation (normalising to the average volume), instead of peak normalisation (normalising to the peak volume) and so allows us to maintain more dynamic range within the programme material. It (inevitably) introduces a new metering system (the LU - loudness unit), and specifies 3 parameters: 1 - Programme loudness and should adhere to -23 LUFS (this is an average loudness signal for the whole programme), 2 - Maximum True Peak measurement of -1 dBTP and 3 - A measurement of the dynamic range of the programme. What is great here, is that we are finally meeting (and indeed metering) to something akin to a loudness level, as well as a peak and a range.

I've been using an EBU R128 meter for about a year now (zplanes excellent PPMulatorXL), and it really has been a joy to use and work with. I still have my conventional BBC PPM meters in my final bounce bus, but now along with the EBU R128 meter. I find that it's only making small differences to how I work - but then I knew (hoped?) that I was producing mixes that were to specification and had  a good dynamic range; if anything at all, I'm more confident now that I can put louder peaks in there.

Ironically, through this investigation into metering, I found that the TP meter that is part of PPMulatorXL was not quite behaving itself (on a Mac running Pro Tools 10), but the guys at zplane are aware of this, have re-created the problem and are fixing it as I type. Thanks to Martin for his support from Zplane. And also huge thanks to the members of IPSNet (from the Institute of Professional Sound) who helped trouble shoot / explain things.

Please note: there's a lot of very techy information about meters - I've done my best to check that I understand things correctly, and indeed then explain it correctly here. So please read around the subject, and follow the links I've used to get to this point.

Right, back to my meters.