Multi-channel what?, you may be thinking. Well, this is an interesting story. Ambisonics is a model for surround sound, developed by Michael Gerzon (British genius, natch) and others from the early 1970s onward. In many ways it’s an extension of Blumlein’s original stereo work from one dimension to three.

Ambisonics has been distinctly unsuccessful as a mass-market proposition. But it has some real advantages over the current surround-sound systems, in recording, production and playback. Key to these is that Ambisonics doesn’t really care how many loudspeakers you have. The distribution format (”B-format”) is a four-channel stream that encodes the soundfield at the listening position and can be decoded to practically any physical speaker layout: stereo, rectangular, 5.1, 7.1, and even 3-D layouts with height (such as a cube or dodecahedron). Or binaural (headphone), with convincing 3D-ness. And doing that decoding well, in a way which gets the most from any particular room, has only really become possible in the last few years with the advent of powerful DSP. In this sense, Ambisonics is a music format which is not only future-proof but also very timely.

At ambisonia.com, Etienne has been building some good things around Ambisonic B-Format recording distribution. He’s well connected in a little community probably very open to Seth Godin’s “music lessons” in a way we don’t see from the mainstream music industry. But of course there’s a big chicken-and-egg problem here. Ambisonic playback on a home system is not the easiest thing to set up.

Now, hopefully I have a few things which can help. I’m fascinated by the potential of Ambisonics and very impressed with the realistic sound of 4-channel playback. I’m a big fan of the Slim Devices products and the way they have been put together; and Squeezebox’s potential as a way to make DSP-processed computer-based music acceptable in a very non-technical environment without any compromises to sound quality. So you’ll see Ambisonic playback capabilities in InguzDSP. But today they’re only capable of two-channel decodes, which kinda misses the excitement of a really immersive surround recording.

There are three ways I can see to fix that. One: support binaural decode, for listening on headphones. I’ve experimented quite a bit with this, and had reasonable success, but not quite enough (limited externalization, slightly odd timbre, and so on) that it’s ready for any sort of wider audience. Two: support direct playback into a multichannel audio system. But this will require either a multichannel Slim box — simply synchronizing multiple players has a couple orders of magnitude too much clock-drift — or a SqueezeSlave capable of writing an abritrary number of channels ASIO, which won’t happen unless I improve my C/linux skills and will in any case be far too geeky a setup for the average user. Third: use the DSP pipeline to encode a multichannel output into AC3, wrap that in S/PDIF stream, and send that to the Squeezebox. If you have a Dolby Digital receiver connected to your Squeezebox digital output: bingo. Not lossless, but decent quality, horizontal surround.

And it’s possible. Quite a lot of work — way more than I can set any sort of schedule around — but something I really do want to do. Because the alternative is to ditch SlimServer and SqueezeBox and Jive and go back to Foobar plus a multichannel soundcard, which would be a big step backward in usability. And after all, usability and the way the audio environment fits into your living room is the sine qua non of success here.

It’ll be an interesting journey, anyway!

So far of course the Ambisonic decodes are all to stereo.  Longer term I’d like to decode to DTS or AC3 (at least for a rectangular layout), since that’ll work “out of the box” on many systems.

To kick off the year, I wrote a small review of Jive on my other blog.

The high-frequency control is pretty much exactly what we asked for.  But the low-frequency control is shallower, more spread out, and ripply too.  This isn’t so much due to the bands being at low frequencies — although that does have some bearing on the problem — but it’s directly related to the bands being very close together.  The frequency scale in the graph (and the way we hear, approximately) is logarithmic, but the processing (FFT) is linear.

So, if closely-spaced bands have this problem, how can it be fixed?  Well, we can build longer filters.  Longer, literally, in time.  Time and frequency are like two sides of the same coin, or like waves and particles…  If we have more time, we can have more exactness about frequency.

The default EQ filters produced by InguzDSP are 4096 samples long (about 1/10th of a second).  This means they don’t use much CPU when being processed, which is handy.

What if that were increased?  Well, here’s another graph.

The green curve is a filter with 4096 samples; cyan is with 8192 samples; and purple is 32768.  Longer filters give a better-controlled curve.

So, in a future release I’ll make the filter length adaptive.  For most people, easily-processed 4096-tap filters are plenty (even “more than enough”).  But if you have closely-spaced bands, I’ll increase the filter length - giving better control where it’s needed.  Unfortunately doing this isn’t completely trivial because my code has some naive assumptions about the “loudness” of a filter.  But that needs a post of its own some time.

“I’m using the parametric EQ stuff, and I want to make a deep notch at 60Hz then a small peak at 70Hz to tame the bass.  But when I do that with the EQ, it doesn’t seem to be as effective as it should be.  Why not?”

To which the answer is twofold: filter length, and shelving filters.

Hmmm.  Let’s start with the “shelving filters” thing.  Filter length (and the little bit of development needed to get longer filters) can wait for another time, since it’s quite an intricate skein to unwind.

Analog EQs are quite a varied bunch (see this great note on the Rane site for some history and deep dive into the various methods).  But the most flexible of the common types is parametric, meaning that you can adjust three parameters of each filter: its center frequency, the amount of boost or cut, and “Q”, or the sharpness of the peak or notch.  With several channels of parametric EQ, you have a really flexible system.

The InguzDSP EQ filters aren’t like that.  They’re “semi-parametric linear-phase shelving filters”.  By “semi-parametric” I mean that you only have two adjustable parameters: frequency and amplitude.  And “shelving” means that the frequency response in between two adjacent bands is a gentle (cosine-based) slope.

Below I’ve charted two three-band EQ filters.  In green: -12dB at 60Hz, 0dB at 960Hz and +12dB at 15kHz.  In purple: +12dB at 60Hz, 0dB at 960Hz and -12dB at 15kHz.  The overall frequency response looks nice; gentle curves between the chosen frequencies.

This design comes from wanting the 2-band EQ  to approximate the classic Baxandall tone control.  Unfortunately there’s a longstanding “bug” — an accidental omission, which I don’t really plan to fix — which means that the two-channel EQ doesn’t “zero the midpoint” (the way a Baxandall is always 0dB at 1kHz).  Instead, two-band EQ is more like the old Quad “tilt” control;  it’s just a single gentle shelf between the bass setting and the treble setting.  If you actually want a Baxandall-a-like control, then use three-band EQ and leave the midpoint at 0dB.

So what happens if you group your chosen frequencies close together?  Well, they just get gentle curves between them.  But, at closely-spaced low frequencies, something else happens: the peaks and dips aren’t as sharp or as strong as you’d hope.  and that’s to do with filter length, which I’ll talk about soon.

For me, a forum and gallery aren’t much use unless I can keep up with them, and for that I use RSS (with BlogLines, mostly).  So here are some feeds:  blog, forums and gallery.

Subscribing is good, but this is a read-write medium we have here, so please do join in.  I’d love to see some other people’s photos in the gallery!

Blog? check. Forum? check. Gallery? check.

Links to all this from the main pages? check.  (That was trickier than I expected).

Meanwhile there’s a really great discussion over in the slim forums.  I hope to dive in there with some pithy and enlightening comments myself, after I’ve ironed out a few wrinkles over here.

The gallery is open - please feel free to post photos of your room, system, (pets, etc fine too).

I’m particularly interested in the system-in-room aspects.  Do you feel your layout is constrained by, um, “domestic considerations”?  (I do… but then, I live in pretty much a studio apartment right now, so there’s not too much real-estate to be divvied up).  Have you tried room treatment (panels, bass traps and so on), and how did they fit in?  Is it really dreamland to hope for an integrated living-and-listening place?  Or, are isolated dedicated rooms are something you have or aspire to?

Some time soon I’l work out how to tweak the gallery to properly support integrated measurement files - real data.  I think the combination of hard data and real-world pictures can help everyone better understand how sound is influenced by surroundings.

If you want to uplaod pix but can’t work out how, do say so.

Happy posting!

Oh yes.  With a photo gallery too.

It’ll be a little while before the look-and-feel is all cleaned up enough to announce to the world.  But if you want to register and join in, that’d be lovely.

“Flatness”.  I’m not sure this is the right name (maybe you have some better suggestions).

I really like a setting of about 7.  The sound opens up a bit - the band steps forward, or grungy bad recordings take on a really decent sound.  6 gets too boomy, sometimes, and 8 is too dry.  (Maybe this number range needs some stretching out.  What do you think?)

Tell me your best “flatness” number.  Do you use it?  At 10?

The “flatness” control works by looking at the room correction filter.  (So it’ll only show up on the EQ menu if you have a room correction filter selected. Just in case you wondered what on earth I was talking about).  The filter is based on measurements in your room, with a measurement microphone.  so “Flatness” looks at the overall frequency response of the correction filter, and inverts that.  Which should pretty much equate to the original measurement frequency response, if the correction filter is equalized to something near flat.

Then it builds a filter for the precentage of flatness you have chosen.  Flatness 10 means skip this altogether.  Flatness 0 means build an inverse which is a smoothed representation of what the room measurement’s frequency response looked like - maybe even a bit more intense in the peaks and valleys, but very smoothed (ERB bands, roughly).  And the range 1 through 9 is all inbetween these two extremes.

So if you’re used to an uncorrected sound with a particular bass lift, or a presence-region dip, or whatever:  adding some of that room sound back, broadly, tends to make the sound more “in the room”, more forward.  After all, I’m assuming you have speakers and a stereo system you kinda like the sound of.

There’s a really practical technical reason to build a flatness filter, too.  Most people aren’t using calibrated measurement microphones.  The off-the-shelf Behringer and so on are reputedly quite good and flat, but are you sure?  What if the microphone has a 5kHz dip - would you ever know?  Any particular sample might be as much as a couple dB different from another, and without calibration there’s no way to correct for that.  So by reducing flatness, we also make a small bet against the microphone’s frequency response, selectively ignorning it a little (but we do trust it’s phase response.  Which is quite reasonable, really).

Anyhow.  Enough rambling.  Have you tried flatness?  Do you like it?

And welcome to the blog, too :-)