Against the Dying of the Light: the Second Cantus CD The Sound part 2
The middle stack, from top to bottom: a Nagra-D open-reel digital recorder (four channels of 24/44.1 data, used to store two channels of 24/88.2 data from the omni pickup); a PrismSound MR-1024T "bit splitter," used to divide two channels of 24/88.2 data from the cardioid pickup into eight channels of 16/44.1 data; a Tascam DA-38 MDM recorder, used to store those eight channels of cardioid data; a pair of two-channel Millennia Media HV-3B low-noise solid-state mike preamplifiers, used for the omnis and cardioids; a Forssell M2a 2-channel low-noise tube preamp, used for the Neumanns and for the binaural DPAs; two two-channel dCS 904 A/D converters, one slaved to the other at 24/88.2; and an Audio Power Industries Power Wedge AC conditioner.
On the right, again from top to bottom: Stax Lambda Pro and Sennheiser HD-580 headphones; the Stax tube headphone amplifier; a Perpetual P-3A D/A converter for monitoring the 88.2kHz data; a dCS 972 format converter to convert the 24/88.2 double-speed AES/EBU link from the dCS 904 ADC to a single-speed 16/44.1 AES/EBU datastream to feed the Panasonic SV-3700 DAT recorder below it. This was used both for the binaural recording and to store 16/44.1 backups of some of the sessions, which I burned to CD-R when I got back to my hotel room.
Each take was slated with a slapstick at the dead center of the stage, this used in post-production as the timing reference to align the pairs of channels. I tried moving the cardioids backward and forward in time in the mixdown, but the theoretically correct alignment was the best-sounding, with a deliciously open, vivid sound. I ended up using the omnis as the main pickup in the mix, with the cardioids added in at -6dB to stabilize the soundstage. The omnis' high end was also shelved down by 3dB to allow the more precise imaging of the cardioid pair to become dominant above 3kHz or so.
With this mix, the lower mids were rather lightweight compared to how they'd sounded in real life, so I used a Z-Systems rdp-1 digital parametric equalizer to provide a slight boost: 3.4dB at 80Hz, with a Q of 2. More than this and the basses began to sound too plummy when the cardioids were added to the omnis. Mixing in the close sound of the Steinway at -12dB in the Debussy proved to be just the right amount with respect to the levels of the vocals.
One problem I hadn't anticipated concerned slight differences in the noise floor for different tracks. The program varied enormously in its dynamics, from the quiet, reflective Barber piece that closes the sequence to the work by Finnish composer Veljo Tormis, which climaxes with the voices being drowned by someone beating the hell out of a large gong. (From the score: "The sound must be so loud that the chorus entrance is not heard.") I thus adjusted the microphone preamp gain for each work during the sessions, to maximize the resolution of each. This meant that I had to adjust the level accordingly for each track during editing and mastering.
However, the ambient background noise had not been consistent, due to wind, noise, and distant traffic. I hadn't thought this would be a problem at the time, because the peak level of this noise was between -57dBFS and -63dBFS, and any changes in the spectra, almost always LF in nature, should not have been audible. Nevertheless, they were audible, resulting in small "gear-changes" at some of the between-track transitions. I eliminated these by creating long crossfades between the songs and, in a couple of cases, some slow changes in low-frequency equalization, applied to just the noise floor and again using the Z-Systems rdp-1.
As all the mixing and equalization was performed in the digital domain, the 24-bit resolution of the original data was preserved.
I downsampled the sampling frequency from 88.2kHz to CD's 44.1kHz with my dCS 972. Once the master had been assembled, I reduced the word length from 24 bits to CD's 16 using the rdp-1's POW-R algorithm, as I'd done for last year's Let Your Voice Be Heard album. This is not noiseshaping per se, but very-narrow-band dither applied just below the Nyquist frequency (half the sampling frequency), conceptually similar to the UV-22 algorithm that used to be available from Apogee.
I am confident that the sound of the CD is true to the sound of Cantus, on four of the coldest days Minnesota was to experience in March 2002. It was a privilege to have been associated with this most moving music.—John Atkinson