Compressing spoken word audio for podcasts

What a terribly lax blogger I've been over the last few months; part of that was due to a couple fo trips to Saudi to finish building a facility in Riyadh (more about that to come; did you know that protective mains earths are not mandated there?!).

Anyway - I look after a couple of podcasts (just spoken word content) and for the longest time I've been using the Conversations Network Levelator to compress the dynamic range of each clip before I import them into Audacity and then edit/normalise them before exporting as an MP3 for upload. Imagine my horror when after switching to a new (old) laptop that I installed Mojave (OS-X 10.14) onto I could no longer use The Levelator; and it's a thing - lots of folks complaining online about it, but it's abandoware so what are you going to do?

 You upgrade the OS and some software stops working...!

 People listen to podcast in noisy environments; in cars, on the bus in earbuds and in other non-ideal listening environments. So, like talk radio the best thing is to compress the signal until there is almost no dynamic range and then normalise it to less than a dB under 0dBfs. Then it's a loud and punchy as it possibly could be and it'll be usable in bad listening situations. The Levelator was brilliant at doing this, but I thought I should at least roll my sleeves up and see how good the compressor in Audacity is, and blow me down it isn't bad!

After a bit of tinkering around with the ratio and threshold I arrived at these values as being best for audio that is peaking around -10dBfs (I'm a broadcast engineer, after all!) - it produces speech that still sounds OK, but has almost no dynamic range!

In this clip the band starts at around 2'15"

Sticking this .WAV file through the compressor set up (above) produced a very usable result which I then normalised before joining it to the rest of the segments of the podcast.

It looks like there is a lot of noise, but it's the sound of a very large room and on the speakers is sounds OK.

Power supply toroid in Mackie SRM450 PA loudspeakers

A friend brought one of these very common PA powered loudspeakers over to the workshop. An internal T4A mains fuse had blown and when I replaced it and re-powered there was a very loud 50hz hum for a few seconds and the fuse in the power cord blew. I assumed it was the audio path that was making the noise and I assumed that one one the probably faults was a short/open de-coupling capacitor that was letting 50hz through to the audio stage.

I found the schematic at Mackie-SRM450 actspk.zip (although this is for the rev.C of the board, but the PSU appears not to have changed).

So - I made sure both fuses were good and with Dave Jones's words "thou shall always check the rails" in my ears I disconnected the output of the large toroid from the two rectifiers/smoothing caps and tried again. I figured I'd make sure the o/p of the transformer was good (there is some de-coupling on the primary). However I got the same loud 50hz for a couple of seconds and one of the fuses failed. The toroid was also noticeably warmer than it had been! So - checking the DC resistance of the primary side showed it to be less than six ohms (so it was pulling an inrush of at least 40A!).

A quick flail around the web showed that the transformer is a known weak point of this design.

 The real bummer is that the transformer isn't stocked by Mackie (or indeed anyone else) and the folks at Save My Light only do a minimum order (ten pieces) from the Chinese factory that wind them when they have enough orders (and the chap there told me he sells an average of two a year).

So, do I just wait or pay him the thousand quid to have ten made?!

Time to keep my eye on eBay for a hopefully working second hand replacement?

UPDATE 16/01/2017: 

So after lots of flailing around the web and finding a few reclaimed ones for more than £150 I came across AJAudio on eBay (Alex Mathew sales@ajaudio.co.uk was super-helpful) and he sold me one for £99.
Job done.

For info - the failed toroid had around 5 ohms DC impedance across the primary whereas the replacement has 8.5 ohms.

It's been shaking the workshop with loud music all afternoon!

The Engineer's Bench podcast - "Top Tips; Audio"

Hugh and Phil talk about some tips and get-out-of-gaol-free cards with respect to broadcast audio.
Go to the website for a PDF of the notes.

https://youtu.be/QJO0Z4DuDFc


Find it on iTunes, vanilla RSS, YouTube or the show notes website.

Audio over cat6a cable?

My podcast-partner in crime, the mighty Hugh Waters asked me how often I run audio signals over twisted-pair data cable in media facility builds. Has has a customer who is eager to do it this way.

I’ve done it a few times and it’s fine with a few considerations.

1. Earthing is still important and since IT people have no idea about proper grounding it can be an issue. If it’s a cat6a install in a TV facility done by me I’d have no worries, but the average IT install might have issues. But as four twisted pairs with individual screens cat6a is ideal for audio. It’s the other parts of the facility I’d worry about,
2. RJ45s don’t have the same DC/LF performance as a good old B-gauge (or Bantam) connector; if the circuits are going to be patched often I’d be wary – mechanically they aren’t great next to traditional audio connectors, 
3. From a wiring perspective; how do you nicely terminate into XLRs from a piece of cat6a?
4. Track shuffling is hard,
5. AES – just fine. Cat6a has an impedance of ~100Ω per pair, ideal for twisted-pair AES. Same observations as 1, 2 & 3.

There you go. I did work at one facility where the engineer had got obsessed with structured cabling and did everything he could over cat6 – analogue video (via baluns), audio, RS422 etc. He liked the idea that you could patch an offline i/o with two RJ45s (stereo i/o on one and video i/o and remote on the other). It didn’t work well and I put in proper cabling and patching after a year.

Sometimes convenience blinds you to fitness for purpose.

AJA HD10AMA analogue audio pinouts

The HD10AMA is a great de/embedder with two HD/SDi outs; very useful gadget. Here is the D25 details.

Managing multiple identical sound devices in OS-X

I use Skype (although I may be looking for alternatives due to Microsoft's proved snooping - see here) and I like to have two sound devices so that the radio can keep playing through my speakers without me having to reach for volume control when I take a call. Also, when podcasting, I use the same laptop to run the presentation, keep the Skype call going and make the recording (that chews up three sound devices!). So along with the laptop's internal sound chip I have two cheap external USB dongles. Since they are identical they show themselves with the same name is all apps and invariably (especially if I've been away from my desk for a day and re-booted the OS without any of the USB devices attached) Skype picks up the wrong sound devices as default. It's trivial to change back but I always get it wrong ("..is the headset the first or second one"?!)

In Utilities is the Audio MIDI setup application (which I've never used before) where you can set "aggregated sound devices" - presumably to allow the same audio to play through several outputs? But - it allows you to create a proxy for a device and give it a sensible name.

So, I made new devices for the two USB sound dongles and gave them sensible names.

This now means that when I look at available sound devices in other apps (particularly Skype) I see things I can distinguish!

iTunes makes broken MP3 files now?!

I've always stuck with MP3 files for music because it is the only format that everything we've ever owned will play. Currently we've all got iPhones and iPods but the car machine is one of those no-name head-units that's a radio and a flash-memory player. In the past we've had an assortment of 'phones and no-brand MP3 players and so I think the choice I made back in the late nineties to start moving all my music to MP3 was valid.

People object to MP3s for one of two reasons;

• It's not an open format like OGG Vorbis - it's notionally "owned" by Technicolor. It is so ubiquitous that I suspect they'd have problems enforcing that.
• It's lossy, and not even the best example of a lossy codec.

As ever Wikipedia has a very comprehensive article. On the first point I think life is too short to get all religious about technology choices. In the case of documents formats - sure; send RTFs rather than DOCXs just for politeness (actually, both formats belong to Microsoft!). Not everyone has the right machine or can afford MS Office (oh, and NEVER send Open Office specific files!).

In terms of the lossy nature of MP3s I'd say that if you encode files well yourself it shouldn't matter for most people and most music. With very little effort you can get MP3s that are so close to the uncompressed data that came off the CD that you'll never know. Audiophiles (who still tolerate all the noise and 2nd order harmonics that come off vinyl - and don't get me started on the RIAA characteristic!) claim that no compression is good, but I suspect they do so for reasons of fashion or self-aggrandisement. The reason I say that is I have actually done the tests!

Back in 1999 I was involved in a project to transfer a large audio sound effects library to a server. The start of the project was to see how well compressed audio was suited to the task. Drives were small and expensive back then and so the success of the project rested one us using compression. So - we compressed several dozen bits of audio to 96, 128, 160, 192, 256kbits/sec;

• Spoken word - properly recorded in a high end audio booth for existing TV voice-overs, male & female.
• Music - a selection of acoustic, classical, rock etc
• Sound effects - from the BBC library, spot effects as well as longer ones (bird song etc)

So these were blind-played to the Oasis Television audio staff (at the time a good example of "golden ears" - people who have been trained to hear audio problems) in properly built audio suites (£10k speaker systems in acoustically dead rooms). So - not amateurs making judgements on sub-£1,000 domestic rigs, but a proper blind-test using professionals.

What we discovered was that past 256kBits/sec nobody could get reliably better than 50% correct - it was as good as if they were guessing which was compressed and which was uncompressed. This seems to fly in the face of current opinion that says that even 320kBit/sec is detectable on iPod earbuds (!) - and don't forget that the LAME and Fraunhofer compressors (reckoned to be the best) have been getting better over the last decade-and-a-half (particularly with respect to VBR encoding).

Lots of people also suggest that other codecs (particularly AAC and WMV) are now better than MP3; I've never been able to hear that when I've compared like-for-like (data rate, VBR vs CBR etc) and since MP3 is so ubiquitous it seems likely that manufacturers would have spent more dollars optimising it that any of those lesser used codecs?

So - I compress my music to 192kBits/sec using the LAME VBR setting and I rarely hear an artefact. There are a few albums I did back in 1999 that I've gone back to because modern compressors are so much better and the little MP3 player I had back then could only hold a complete CD at 128kBits! The cruel irony is now that I know what I'm listening for and can (just about!) afford decent speakers my hearing response is rolling off quite markedly. Pretty soon AM radio will sound good. I've also found that when I mix live music I drive the high-end a lot more than I used to and that must be the same effect.

There is another effect that people talk about - how tired you get listening to compressed audio - the brain doesn't like artefacts that you don't encounter in nature. I think this is true, but the people who make play of it tend to be vinyl & FM radio fans - both of which are covered in very unnatural artefacts.

So - to the point of the post; I discovered that a couple of CDs encoded by iTunes over the last year wouldn't play off a USB stick in the car. I had to re-encode them on the old AltoMP3 maker software I used to use. flailing around online seems to suggest it's the way Apple sticks artwork in.

AES audio on D-25 connectors

I've been working at a facility that delivers DCP masters to cinemas and the thing that we had to pay lots of attention to is the pinouts for various multi-channel audio servers and monitoring boxes, principally;

• Dolby 650 & 750 surround processors - the gadgets that "tame" a room to make it sound like a cinema or screening environment should
• TC Electronic TM09 multi-channel monitors; used for loudness monitoring (R128 & ITU.1770)
• Dolby DS100 & 200 servers
• Doremi DCP2000 servers

They are all different!

I could just list all the pinouts from the manuals but here are a couple of grabs from cable schedules that show exactly how to do it over DMP-10 cable, krone blocks etc.

Odyssey Products have a great PDF summerising them

 

Passive low-value audio pads

I've often had to knock up audio attenuators to make music gear (which is typically +4dBu for zero level against 0dBu for broadcast) and my usual m.o. is to approximate everything around a 10k potentiometer; typically 100ohm sending impedance, 10k ohm terminating impedance and an H-network for balanced lines and a T-network for unbalanced. You can find numerous examples online.

So - all credit to my colleague Matt for saying "..no, no - let's do it properly";

 

So, for a variable 1.5 -> infinite pad you need Z1 and Z2 at 390 ohms and a 5K potentiometer. 

Blow me down, the 5k pots arrive with their wipers at the centre position and all six that I've made so far have been bang on 2dBs at that centre point. It pays to be precise.

Beyerdynamic DT-250 series headsets

I'm just prep'ing a few Beyerdynamic DT250 headsets for a customer - football club we work with a lot! Their commentary guys love these headsets and the quality/comfort is hard to beat for either pitch-side or commentary booth use. They come with a proprietary connector and being German it's all beautifully manufactured.

This customer sometimes uses these with portable Glensound ISDN codecs (reporter-type mobile packs) and sometimes with Yamaha audio mixers and so they need both the 5-pin XLR option and the 3-pin XLR/1/4" stereo jack cable. Since removing the rigged cable requires a screwdriver and is fiddly I normally ship these with the 5-pin cable and make up a breakout;

Alternatively you can order the two different cable options, if you can get your colleagues in Sales to pay attention!

Electrical Network Frequency analysis

I've been running a training course at the Metropolitan Police's Video and Audio Forensic facility in Sydenham. I've been able to chat to some of the technical forensic guys and one thing has blown me away - Electrical Network Frequency analysis. 

The whole of the UK mainland is on an electrical power-grid; as generators run up and connect to the grid they have to be at the same frequency and phase-locked to the 50hz AC supply of the rest of the country (or there would be sparks!). Now although the mains supply is pretty accurate at 50hz load variations cause momentary changes that are reflected across the whole country; typically less than 0.2hz either way - mains can be (at any moment) 49.8 through to 50.2hz, but even over short integration periods it is darn close to fifty. I couldn't find any info for the UK, but here are some traces taken from the mains supply in Romania in 1998, three towns, separated by 800Km. 

Notice how the waveforms track each other precisely. It turns out that the pseudo-random sequence is very identifiable after the event. For this reason the MET have been sampling it for the last five years. They know exactly what the mains in mainland UK was doing at any point since 2007. It also turns out that induced mains hum is present in most audio recordings; either through pick-up in the power supply or from the hum of lightbulbs etc. You have to work hard to make an audio recording that doesn't have this watermark down there in the noise. 

They've even discovered they can extract it from the regular flicker of lights on video-recordings. This means that you have a forensic tool for checking the record date and time on evidence. 

There is a Wikipedia page on the subject and they've challenged me to send them some samples of speech recordings I've made over the years to see if they can time-match them.

Audio podcast 2 - the engineering!

Hugh and I continue our discussion of audio and make particular mention of cabling for TV facilities.
Find it on iTunes, vanilla RSS, YouTube or the show notes website.

The next podcast - Audio 101

Hugh and I go over the fundementals of sound and recorded audio in the first of a two-parter Find it on iTunes, vanilla RSS, YouTube or the show notes website.

I'm representin' at BVE

Today I'm delivering a 'training taster' session at BVE 2012. Oh, the title link is my notes if you want to download them.

People's expense accounts depend on their unquantifiable skills!

In 1999 the Super Audio CD format was released - higher sampling rate and longer word-length than the venerable 44.1Khz/16-bit Red Book standard that traces it lineage back to the late seventies and the Sony F1 digital audio system.
I've spoken to audio engineers who have made a very good career out of there being a benefit in re-mastering recordings to this newer standard. Their contention is that the difference is "night and day" (please go back and read that post).
Anyway - in 2007 a couple of chaps from the AES did a double-blind test to see if audio professionals could tell the difference - it turns out they can do no better than random. Remember - that was audio engineers, dubbing mixers, and other people who know what to listen for in properly recorded audio. Mix Magazine did a very good write-up under the title of The Emperor's New Sampling Rate!

This all reminded me of a project I was involved in at Oasis TV in the late nineties where we were home-brewing an audio-FX server for the dubbing suites. At the time 9 gigabyte SCSI drives were £1,500 and so compression was implied! None of the dubbing mixers liked this idea and so I made up a CD of various recordings; spot-effects, different music styles, dry vocal recordings and finished mixed programme. The compression we were using was MP2 (so not as good as the now-ubiquitous MP3) at 128, 164, and 192 kBits per sec (as well as uncompressed).
Remember - these were the golden-ears listening on £10k matched amp/speaker combos. It turns out that somewhere between 164 and 192kBits per sec these guys dropped to about 50% accuracy in discerning the compressed audio from the original.
Actually I think it's a bit more complicated than what these two double-blind tests suggest; I store all my music at 192kbit MP3 encoded using LAME 3.9 - for 99% of my music I can't hear the difference. However;

• On some passages (typ. splash cymbals and some acoustic guitar parts) I am aware of compression artefact's.
• An old VT editor once told me (around fifteen years ago) that although he liked the look of (the then new) DV format he felt more tired after a day of editing DV footage compared to BetaSP - the differences aren't immediately clear but over time one is better (in some way?) than the other.

I do believe that you can only get to the truth of these things by statistical analysis - I place no faith in audio professionals who expect their view to be taken seriously without the numbers to back it up. Their salaries depend on them being able to 'hear' the differences - if they are there or not.

Monitoring in 5.1

Nice article that Colin Birch (@InformedSauce on Twitter) mentioned this morning - I PDF'ed it as the online reader that Broadcast Engineering World use is clumsy. Click the link in the title of this entry.

Rio / Sonic Blue / Dell Digital audio receiver

I picked up these on eBay for a fiver a go! They are tiny little Linux boxes that stream audio from a nominated PC on your network. The original Rio/Dell software is very 2002 and although adequate offers no integration with iTunes or Windows Media Player. When you power the box it shouts out a BootP request (that's an old protocol I haven't had to deal with since Phillips/Thompson routers of the mid-90s!) and that starts the download process of a tarball that has the Linux kernel to run the box. Once that's loaded it grabs an IP address via DHCP (from the same client software that serviced the BootP request - I guess not many people had home-routers when this was a product!) and you can then stream your MP3s to it. It has a nice display, IR remote and the audio output quality is good.
However, since it's demise it's become a target for home-brewers and hackers and there is a new kernel that makes it look like a Squeeze Player - the standard that Logitech have used for all of their home entertainment products. This makes it quite an interesting gadget as a network music/streamed radio player and it didn't take me very long to get it working with my Windows 7 media center. Using Logitech's server mean it integrates brilliantly with iTunes and you get access to all the BBC's iPlayer and streamed radio.


Once you've got Squeezebox running you can use it to stream to WinAmp, your iPhone or even another instance of iTunes (rather perversely!).

So, here are the bits you need;
Rio's v. 1.04beta which was the last release. It's the only one that works with Windows 7 and it doesn't have all the instabilities previous releases had with 10 vs 100BaseT ethernet (this was the early noughties after all!).
SlimRIO is the tar-ball with the new kernel to make the box behave like a Logitech Squeeze player.
Squeezebox is the server that makes it all work nicely.
Finally you'll probably need a copy of the LAME encoder so that the server can transcode AAC, RealMedia etc into MP3 streams for live radio - lame.exe - it needs to go in;
\Program Files\Squeezebox\server\bin

Tektronix and audio loudness measurement

I just had a rather splendid lunchtime presentation from Tek regarding the new firmware for WFM & WVR-series test sets. EBU rec 1770 has been around for some years but a couple of things have stopped it's widespread adoption.

• It's integration time for short-period measuring is three seconds - Channel Four (who previously were the only UK broadcaster who got shirty about perceived loudness) always specified a Chromatek meter which used a four second rolling window.
  
• It's long been understood that most archive material fails 1770 in it's original state but the inclusion of a silence gate mitigates this.

It seems like whole industry is tip-toeing around the dirty little secret that commercials producers mix audio with a very limited dynamic range so as to make them more punchy. It's in their interest (and the broadcasters who make their living out of them) to not embrace this. It's why my Mum complains to me about how loud the ad breaks are. The EBU should stop pretending this is about programmes, it's about commercials and the sooner they enforce loudness limits the better!
We got to have a play with the new Tek firmware and they have done an excellent job of interpreting the LUFS scale. They make it very easy for an operator to see where a programme is and if the Dolby DialNorm (dialogue) and Dynamic Range figures match what is measures.
More when I've got a copy to put into my WFM 7120.

Why did late 60's/early 70's rock singers have two mics taped together?

I've been enjoying "Guitar Heroes at the BBC" on BBC4 where they compile clips from Whistle Test, Rock goes to College, TOTP etc. I've always wondered why rock singers from a period of only a few years would have two mics taped together. By the time I was paying attention in the late 70's the practice seemed to have stopped so I suppose it was a technical development that made the change.
I asked the question on Twitter and Facebook and got great rock'n'roll answers; "...so they could take it to 11", "early form of stereo recording" etc. In fact when I went back over my old BBC notes I had been told why they did it but only a few weeks out of university I don't think I understood common mode rejection!

Ronnie Van Zant of Lynyrd Skynyrd c.1973 - two mics!

So - having re-read my notes and had a trawl around the web (my word, there is some awful rot spoken by people who know very little!) here are the two reasons (and I'll list them based on the technology that fixed the problem), they both rely on the fact that the two mics are wired anti-phase to each other and the assumption is the singer sings predominately into only one of them (doesn't matter which).

1. pre-compressor/limiters you needed a way of loosing some of the induced stage and line noise - this does it.

2. pre-parametric eq - you needed a way to reject howl-round and this does it.

So - you mix the anti-phase feeds in two channels on the desk and all noise/feedback etc gets canceled and the voice (predominantly coming down one feed) remains. Interestingly another technique to gate a mic is to have either an optical detector on the mic stand or a pressure mat in front of the mic which mutes the channel when nobody is near the microphone.

Loudness Measurement

This how to guide highlights the benefits of the Loudness Measurements software on the WFM6000/7000 series and WVR6000/7000 series products.
Until NAB this year (when Tek introduced this update) the predominant way to measure perceived audio loudness was with some propriety instrument like the Chromatec (approved by Channel Four). Tek have done the right thing and not only followed Ofcomm's new guidelines but they've devised a scale that is as easy to read/understand as a PPM.

Hopefully they'll do the same thing with PSE measurement and put Harding out of business!