CA-72 vs. MODEL D

The three sources

MODEL D hardware

A Behringer MODEL D, recorded at 48 kHz and 24 bits through an audio interface. Each patch was set by hand, photographed, and played from MIDI with timed notes. Its own background noise is left as recorded.

CA-72 0.2.8

The public 0.2.8 release, the unmodified plug-in binary, rendered offline with the same saved control values as 0.3.0. It was not fitted to the recordings separately, so it shows where the release stands with these settings, not its best possible match.

CA-72 0.3.0

The public 0.3.0 release, whose oscillator waveforms were calibrated against this hardware. For each example its front-panel controls were adjusted to match the recording, then the settings were saved; the audio is an offline render of exactly those settings, and the screenshot is the plug-in showing them. ENTROPY, CA-72’s per-note variation, is off in every example. The clips were rendered from 0.3.0’s development commits before its release; the released plug-in, rendering the same saved settings, reproduces every one of them, at every QUALITY, sample for sample.

What was done to the audio

Each source has one fixed playback gain, chosen so the notes play at the same loudness, and one fixed start point, so the notes line up. That is all: no EQ, filtering, compression, pitch correction or time stretching. Every clip is mono and plays centred. The CA-72 renders its two output channels, which are averaged. The files here are lossless FLAC and decode to exactly the published samples. Each example’s gains, start points and checksums are listed under its “Levels, measurements and provenance”. In the four feedback examples, the gain is set from the two lower notes only.

Every example can also be heard at each of CA-72’s QUALITY settings, LO, HI and ULTRA, from the same saved settings with only QUALITY changed. The 0.3.0 clip is the one the example was matched at: HI, or ULTRA for feedback. The other two keep its crop and gain, so any change in level is heard as rendered.

The two-oscillator bass and the detuned lead have one addition. Like the hardware’s, their two oscillators run freely and drift in and out of phase, so the sound swells and changes colour from note to note. Their starting phase can’t be set, so each CA-72 render starts a moment early (4.25 s for the bass, 0.23 s for the lead), which lines its beating up with the recording’s. That lead-in is cut off and never heard. The glide example uses a similar silent lead-in, so that CA-72 starts from the same remembered pitch as the hardware. Both examples are their six-note phrases; the separate held notes recorded for measurement aren’t published.

The comparison clips here have nothing added. The blind test uses separate copies in which recorded hardware noise is added to the CA-72 side, so that the quiet background at the start and end of a clip can’t identify it. To hear the instruments as they are, listen here.

What the graphs show

The graphs are measured from these same playback files. Depending on the example they show the spectrum, the level over time, individual harmonics, pitch, a short stretch of waveform, or the spectral peak through a filter sweep. They show the two sources being compared. Hollow points mark values too close to the noise to trust.

Where they still differ

• The filter and envelopes are close but not identical, and the difference grows with emphasis. It is clearest in the 30-second sweep at emphasis 10.
• The oscillators run freely on both instruments, so beating between them and the waveform phases are not the same from one recording to another.
• At high resonance, some pitch estimates late in a held note are unresolved, because the filter’s own ringing competes with the oscillator.
• With one glide setting, the shape of the glide and the octave tracking differ slightly from the hardware.
• In the detuned lead, the hardware’s second oscillator tracks the keyboard slightly less steeply than its first, so their beat slows a little on higher notes. CA-72 has one detune for both, set to suit the phrase as a whole.
• The feedback examples play A, where both loops lock to a tone near 55 Hz. On some other notes, such as C and D, the hardware’s loop pulls to a fraction of the played note while CA-72’s stays near 55 Hz, so they sound further apart.
• Released 0.2.8 shares 0.3.0’s control values rather than being fitted on its own, so some of its differences come from settings, not only from the engine.

We don’t reduce these comparisons to a similarity score. Listen, and look at the measurements.

MODEL D is a Behringer product. Idle Foundry is not affiliated with Behringer.