Filter Delay: Arrange Three Distinct Replies

Read the routing before touching Pan

The three lines do not receive identical signals. Line 1 takes the left input, line 2 takes the sum of left and right, and line 3 takes the right input. Each then has its own filter, delay, feedback, volume and pan. Pan moves a line after processing; it does not select which input channel feeds it. That distinction matters particularly with stereo material.

The original eight-second stereo test phrase below gives those routes something to work with. Free times are 180, 310 and 460 ms. The corresponding filter centres are 700 Hz, 1.8 kHz and 3.2 kHz. Lines 1 and 3 sit partly left and right at −9 dB; the summed middle line is quieter at −15 dB. Dry is 70%, retaining a direct reference.

Build it again by subtraction

Switch lines 2 and 3 off and follow the first reply. Change line 1 Pan toward the centre. You move the same left-fed reply, not a new stereo mixture. Bring line 3 back and hear its later, brighter response. Finally enable line 2. If that addition mainly makes everything louder or busier, lower its Vol rather than forcing it to be as prominent as the outside lines.

The middle line sums both inputs without a compensating half-gain stage. Correlated left/right material can therefore feed it more strongly than either outside line. Its lower starting Vol is deliberate. These controls add wet lines alongside the Dry path; Dry is not a complementary Dry/Wet crossfade. Turning Dry down does not automatically turn the echoes up.

Try setting line 3 to 360 ms instead of 460. Does its reply become a useful second accent or collide with an existing tap? Restore 460 ms if the pattern loses its separation. For tempo-based work, enable Sync on each desired line and use the sixteenth counts: three, four and six make a different arrangement from three equal counts. A millisecond edit is irrelevant while that line is synced.

Texture comes from repeated filtering

Strip the arrangement down to one centre line with the synthetic noise bursts below. At 320 ms, Feedback 45%, Freq 1.8 kHz and Width 2, there is enough spectral information to hear what the band removes. Noise is useful here because it does not ask you to infer filtering from just one pitched note.

Compare Width 2 with Width 5. This width is measured in octaves: at 1.8 kHz, Width 2 places nominal filter edges at 900 Hz and 3.6 kHz. Wider settings pass more low and high content. The implementation uses a high-pass and low-pass pair, so these are gradual edges, not brick walls. Listen for the later bursts becoming more band-limited as feedback sends them through again.

Reduce Vol by 6 dB while keeping Feedback at 45%. The whole trail gets quieter, but the feedback ratio is unchanged because the loop is tapped before Vol. Now restore Vol and lower Feedback to 20%: later repeats fall away faster. Use the first repair for an over-loud texture, the second for an over-long one. If the texture needs extreme gain to remain audible, widen the filter or choose another centre frequency before adding more feedback.