Groovebox and synthesizer audio input latency measurements (passthrough latency).
Out of curiosity I did some latency measurements which might be of interest to others. How long does an audio signal take to go through a groovebox or synthesizer input to output?| Device | Latency [ms] | Comment |
| Novation Bass Station 2 | near zero | analogue |
| Roland Aira J-6 | near zero | Mix In (analogue mix?) |
| Elektron Syntakt | < 0.3 | impressively low latency |
| Source Audio Atlas Compressor Pedal | 0.6 | plus optional lookahead time 0-2ms |
| Roland MC 707 | 1.3 | |
| Roland SP404 MK2 | 1.5 / 2 | depending on settings (direct / filter, isolator, input compressor) |
| Korg NTS-1 MK2 | 2.9 | |
| Access Virus TI2 | 6.3 | |
| Novation Circuit Rhythm | 11.1 | even higher with some Grid FX, 27ms with LoFi effect |
| Novation Circuit Tracks | 11.4 | unfortunately a bit high |
Opinion
For comparison, at 125bpm a microstep on the Circuit Track's sequencer is 20ms in length which makes an audible difference. Audio input latency usually is not a problem for pads and drones. For drums I try to stay below 5ms. MIDI can also add a millisecond or so of latency (not measured here so far).The Circuit Tracks is my favorite groovebox but I mostly don't use the audio inputs any more because of latency. Instead I either route the Circuit's audio through a paired synth in a couch setup (NTS-1) or go through a mixer.
Measurement
Measurement was done with a very short square wave sound from Novation's Bass Station 2, splitting the mono output to stereo. One channel goes directly into a USB audio interface, the other channel goes through the synth to be measured and then into the other stereo channel of the audio interface. To make sure there is no additional unexpected latency, stereo channels of the output signal were switched for comparison. Audio is recorded with Audacity and the offset between the two stereo channels is measured using the little plugin show-duration (credits to Steve Daulton).
Precision
Precision of measurements should be about +- 0.2ms. Theoretical accuracy is about 50 microseconds (2 * 1/44100Hz * 1000ms/s ≈ 0.05ms) but I might also be off a couple of samples selecting the peak. Results are generally quite reproducible.