Better Recordings of Sermons

30 08 2007

I’ve been thinking about the recording of messages lately. I’ve been asked about it a few times in the last few months. The question usually goes something like this, “We want to record our pastor’s sermon, should we go straight to a CD recorder, or into a computer then burn a CD?” If you have read this blog for any length of time, you’re probably thinking my answer is, “It depends.” But you’d be wrong. In this case, I always like to go to a computer for recording. The reason is simple: When you record straight to a CD, what you record is what you get. If you start to early, you have a bunch of dead time up front. Stop to late, same thing at the end. If your level was too low, the CD will be too quiet. Too loud…well, even the computer won’t save you there.

So I’m going to tell you what I do here at Crosswinds. This is not necessarily the definitive way of recording a message. But it works really well for us. We have dual destinations for the recording, CD duplication and the web. My goal for the finished product is a message that is easy to listen to, without a lot of intervention (ie. adjusting the volume up and down) on the part of the user. I consider the environment and equipment people will use to listen to the message–it will be either their car or at the computer. Not exactly the greatest places to discern maximum quality. That’s why I go for listenable. Yeah, I know that’s not a word, but work with me here, OK?

Signal routing wise, we take the direct out of the preacher’s microphone and run it into a compressor. You should know that the pastor’s channel is already insert compressed, albeit pretty mildly, to even out the volume in the room. Yes, I know this is double compression and I could eliminate it if I double bussed the channel. For what we’re doing it’s not worth it. From the compressor, we run into a 3rd party sound card (way, way better than on-board audio). You could also use a USB or FireWire audio interface. While recording, we take care to keep the peaks at about -12 dB to ensure we don’t run out of bits and distort the signal. Finally once the message is recorded, I apply some additional dynamics control to it. Yup, that’s right a third pass of compression.

At this point, audio purists are tearing their robes, putting ashes on their heads and crying, “Oh, the humanity!” I don’t care. I want a CD that I can put into my somewhat loud truck’s in-dash and listen to it while driving down the road without turning the volume up every time the pastor gets quiet.

I thought some visual aids might be helpful to explain some of what we do. The following example was taken from another churches website to demonstrate via waveforms what we’re doing. I’m guessing the recording was not compressed in any way prior to being recorded. In my setup, we already have a mild 2:1 comp, plus a little more aggressive 3.5:1 before we get to the recording. But this shows what you can do “in post” when you record to a computer (or if you’re just a purist and want to jump through whatever hoops you need to to only compress once–knock yourself out).

Let me issue a disclaimer here: I am going to discuss ratios of Loud to Quiet in a minute. I determined the ratios based on pixel counts of the waveforms. I know they’re not accurate dB ratios and do not represent true loudness levels. But for the purposes of illustration, work with me here. We’re going for concept, not 100% theoretical accuracy.

Let’s look at the original waveform, as recorded.

The Initial Waveform

As you can see, there is a pretty wide dynamic range to this recording. The red bar represents the volume of the loud parts; the yellow, the soft. It was also recorded pretty low, so I really had to crank it up to hear it. What we’re seeing is that the loudest parts of this passage are roughly 13 times louder than the quiet parts (refer to above disclaimer…). In practice, that burst at the beginning (on the left) was really loud, but by the time we got to the right, I was having trouble hearing it over the noise of the fan in my room. I had to keep turning it up.

So let’s see what happens if we were to apply a compressor filter to this in our favorite audio editing program (for this example I used Audacity, which is really cool, really powerful and really free). To start, I tried a compressor with the threshold set at -30 dB and a 3:1 ratio. This is how it looked afterward.

-30 dB, 3:1 Ratio

1 Compressor Was Applied

Look at the difference. Now our Loud to Soft ratio is somewhere around 5.5:1 (again, see above disclaimer…). What this means is that there is significantly less difference between the softest passages and the loudest ones. The overall level is pretty low however, so we’ll apply a normalizing filter to it. After we do that, it looks like this:

Normalizer Dialog

I will typically normalize to -1 dB, just to give it a little buffer (which, by the way, is why I didn’t normalize to 0 dB in the compressor dialog). The normalization process takes a look at the signal and applies gain to the entire recording until the highest peaks touch the level you specify (the maximum amplitude). We do this at this point because we know what our highest peaks are (unlike when we are recording live), and we may as well take advantage of all the headroom the system has available.

1 Compressor, normalized

That looks better, but when I played it back, there was still a little more dynamic range than I wanted. So I got a little more aggressive with the compressor. This is how it looks with the threshold set at -35 dB and a 4:1 ratio

-35 dB, 4:1

1 Compressor is applied

Ok, now we’re getting somewhere. What we are doing here is taking the loudest parts of the message and bringing them down closer to the softer parts. In this case we’re down to just over a 3.5:1 ratio between the loud and soft. Again, we’ll hit it with a normalizing filter and it looks like this:

1 with a normalzier

Now that’s what I’m talking about! That will be super easy to listen to, and believe it or not, there’s still plenty of dynamic range in the recording to easily tell when the speaker is emphasizing his words and when he’s pulling back and speaking softly. The implied dynamics are there, but now I can actually hear what he’s saying.

If this was recorded at the proper level (-12 dB peaks), a -35 dB threshold would have been way too low, and would likely have sucked the life out of the recording. You don’t want all the peaks at exactly the same level, because that just sounds weird. You will have to experiment with this to determine what sounds good for your system.

What is the lesson here? Whether you want to compress the recording on the way to the computer is up to you (though I think every speaking mic should be insert compressed for better performance live anyway), but the above examples give you a good idea of what you can do to maximize the listening pleasure of the audience after the fact. I have two presets set up in Audition (the software we use at church to record with) that first compress then normalize the signal. We save a .wav file, which is burned to a CD, then an MP3 which goes on the web. Since we started doing this, we’ve not had a single complaint about the volume level of the sermons on the CDs or on the web.

And again, I know I’m taking liberties with the concept of dynamic range here, and I know this is not, purely speaking, the most pristine way to record audio. So no flaming comments about the inaccuracy of my math or the flaws in my signal chain, OK?

Given the time constraints we’re under to turn these recordings around, and the fact that the FOH engineer has enough to do mixing the service, never mind constantly tweaking recording settings, this works really well for us.

Try applying some post-recording compression this weekend. Best of all, if you don’t like it, there’s always “Undo!”





Settings for Good Vocals

29 08 2007

Dave has an excellent post on the effects, EQ and settings he uses for vocals. Much of it applies to the Digidesign Venue he gets to use (lucky dog!), but he has a great section for those of us stuck in analog-land. If you’re not reading Going to 11, you really should be. Really.





Mixing Guitars

23 07 2007

Tim Corder found a good article in Mix magazine on mixing guitars. While Mix focuses on mixing in the studio, many of the concepts apply to us live guys as well. Check it out here…

Mixing Guitars





Audio for Video: Using High Pass Filters

23 07 2007

Note: This blog has moved to a new home. Church Tech Arts now has it’s own domain; you can view this post,
along with hundreds of others at the new location. Thanks for reading!

Let’s admit it: It happens a lot. Even when you pay careful attention to the audio you record for a video, and you used a good mic (you did use a good mic, right? if not read this…), you can still end up with a bunch of background rumble and noise in your recording. It happened just the other day to someone at the video production company I work for. They were shooting in a grocery store, capturing some interviews. They used a good shotgun mic, with good directivity to cut down on the ambient noise. However, there were those dreaded coolers all over the store, and if you listen carefully (mainly because our brain normally tunes them out), you’ll hear compressors running. Back in the studio, it sounds like a truck going by the entire interview.

Because it’s a complex noise source, trying to run a noise reduction program on it probably won’t work well (and even when the noise goes away, it is often replaced by unwanted digital artifacts of the FFT process used to perform the noise reduction—but that’s another post). However, we do have one tool in our utility belt that can help (actually two, I’ll get to the second, which should actually be the first, in a second): Enter the high pass filter.

A high pass filter is just what it sounds like—it lets high frequencies pass, while blocking low frequencies. Super-basic HPFs are a simple on and off switch with a pivot frequency (the frequency at which it “passes” signal) and slope (how quickly it drops off the signal below the “pass” frequency) set at the factory. Better HPFs that come with higher level editors like Premier Pro and Final Cut allow you to select the pivot frequency. Here is an example of an HPF with a pivot frequency of 120 Hz, and a slope of 12 dB per octave (that is, at the frequency 1 octave below the pivot frequency—60 Hz—the level will have been reduced by 12 dB).

High Pass Filter Example Graph

You can see how the frequencies above the pivot frequency pass by unaffected, while the ones below get rolled off pretty quickly.

This is when it actually gets useful. For the male voice, the fundamental frequency of the lowest notes one speaks is between 85-155 Hz. For a female, it’s a little higher, perhaps 165-225 Hz. This means that there is no real information that we need below 85 Hz for males and 165 Hz for females. And in reality, because of the way we hear and the way the voice is produced, there are plenty of harmonic frequencies that our brain will interpret clearly to make up for missing fundamentals.

So let’s say we have a compressor running in the background of a female interview. We can safely dial up a HPF with a pivot frequency of 165 and not loose any of her voice. We can take it up even higher to eliminate more of the noise, and the clarity will improve markedly. In fact, the voice will “sound” louder once the low frequency stuff is removed because we can hear it better.

So this is exactly what we did for grocery store woman. We dialed up an HPF with a pivot at around 150 Hz, and it totally transformed the audio. There was still some higher frequency noise, and it was obvious she was standing in the store and not a studio, but the clarity of here voice was improved substantially. If I have time later this week, perhaps I’ll grab a before and after sample and put it up here.

Earlier I mentioned we actually have 2 tools in our tool belt. The other one may be on the mic itself. Many professional shotgun mics (and some interview mics, and the occasional lapel mic) have a HPF built in. For example, my beloved Audio Technica 835B has a switchable roll off at 180 Hz at 12 dB per octave. That means at the lowest fundamental of a male voice the mic will be 12 dB down, which is generally not a big deal unless you’re interviewing James Earle Jones. Normally, I like to leave this switched on because it eliminates a lot of room rumble, AC noise and other nasties right at the source. It’s just a good idea. If you use this when you shoot, you will require less processing in the edit suite.

Of course, you’ll want to listen to it through some good headphones first to make sure you’re happy with the sound. You do have good headphones, right?