While working I’ve been listening to music on my Mac on my ER4s. Which is nice, but I don’t like wearing headphones all day. I didn’t want to go out and buy a set of high-end ‘PC speakers’, although something like the Fujitsu TEN might be fun for an afternoon. And anyway, since I bought my B1s, I’ve had a pair of decent speakers spare — either a pair of Neutron 3s, or Dynaudio Contour 1.1s, depending on which I keep as rears on my main system.

So I decided to build an amp instead. Looked briefly at a few cheap tube kits, and decided I didn’t want the hassle or sheer bulk of something with a huge transformer — or the risk of the cats getting their paws somewhere high voltage!

After some research, I’ve ended up intrigued by the whole chipamp movement — the promise being ‘around 10W of high-end sound for under a couple of hundred dollars and some DIY’. Wanting a nice enclosure, I ended up ordering a Charlize Tripath module and one of Art Taylor’s custom cases. Put it all together one morning a couple of weeks ago, powered from an external power supply.

How does it sound? Well, it’s low power, so efficient speakers help. I’ve never really liked the Neutrons much anyway, and they sound boxy on the Charlize. The Dynaudios sound as good as they ever did (which is pretty good for their size) — and it’s them that have stayed hooked up to the Mac. But the real test for my little chipamp was the B1s, driven from either my homebuilt FLAC jukebox or by my Arcam FMJ DV27. And the truth is that it actually sounds quite amazing, for size and price! I need to spend more time going through my music, but — especially for recordings of vocals and acoustic strings and percussion — it does as well, to my ears, as my Bryston – and possibly better (!) when it comes to depth and positioning. The percussion on No Lonely Nights, from Keith Jarrett’s Rarum, in particular, sounds astonishing. Likewise Cat Power’s Metal Heart, off Moon Pix.

Big, dense, heavily studio-processed recordings don’t fare so well, muddying up a bit. But. This is from an amp the size of a CD case, that cost me maybe £150 to build. I could live with it, happily, as long as I had a great pair of speakers to hook it up to. Impressed.

Have updated my music player to use the JACK low-latency realtime library for output, and replaced Mplayer with alsaplayer as the media player. Have also dropped the Meridian 518 and Yamaha decoder from the output chain, feeding audio straight into my Bryston via the music box’s onboard DACs.

Given the hardware (and purpose of the box), there’s little benefit to be had from the low-latency features of JACK, but the realtime support helps keep the output smooth in the face of other stuff running on the box (CD ripping, for example).

And something in the combination of the simplified signal path and JACK/alsaplayer has definitely tightened up the high end a bit more, which is always a good thing. If only I had some Vivids to audition this all properly!

Now what I really want to play with is to tweak JACK to emulate the 518’s noise shaping/upsampling algorithms. Later…

Thinking more about glanceful interfaces, and the communication of complex multivariate datasets. For reasons I haven’t gotten around to writing about here yet, I’m veering towards sound cues for a lot of things, particularly binaurally-located vocal cues. I’m looking for a pipelining spatialiser using some simple head-related transfer function (HRTF) that I can feed audio into at approximately realtime. For ‘earcons’, simple samples are easy. For vocal cues, I’m thinking of using FESTIVAL. But I need the spatialiser, and I can’t find one that runs on Linux and accepts a stream input. Maybe I’ll have to hack something in Max. I’d rather not.