Parallel processing: check latency, phase and blend level

Map the split, each branch, and the point where they rejoin. For a return meant to carry an effect alone, confirm that its dry/wet control is fully wet. A parallel compressor's processed output will still resemble the original waveform; that does not mean the plugin is adding an unwanted dry copy.

Polarity inversion changes the waveform's sign. Delay shifts it in time. A filter can change the phase of different frequencies by different amounts. Flipping polarity cannot align a delayed branch or undo an EQ's frequency-dependent phase response.

Delay compensation can align reported plugin latency, but monitoring and routing exceptions matter. Ableton documents reduced-latency monitoring and return-routing cases separately. Consult the equivalent host guidance rather than assuming a working serial insert proves a parallel route is aligned.

Use the same transient-rich loop on both branches with identical gain and panning, no modulation, and no remaining processing. Two identical aligned copies summed at equal gain produce approximately 6 dB more level than one copy. Inverting one should cancel them. These are the ideal test conditions, not an expected result from two compressed or distorted signals.

If they do not cancel, check one cause at a time: branch level, panning, different source taps, clip stretching, and delay compensation. Restore normal polarity afterward. Investigate a timing offset before moving clips manually, because a quality-mode or routing change can invalidate that correction.

Bring the intended effect back, then blend at matched overall listening level. If hollowness starts with an EQ on one branch, compare less EQ, another placement, or a documented phase mode. Linear-phase EQ may address that interaction but adds latency and can introduce pre-ringing; it is not a general quality upgrade.

Live documents a compensation exception when a return feeds an audio track whose corresponding send remains active. For example, if Return A feeds Audio Track 1, right-click Track 1's Send A knob and choose Disable Send. Merely turning the knob down does not disable the route. Apply this fix only to that documented case.

Render a short passage and import it without unintended stretching. Compare the export with live playback and the summed mono signal. Recheck after lookahead, oversampling, monitoring, or routing changes. For external hardware, measure the round trip with the host's procedure rather than treating interface buffer size as the complete delay.

Keep the parallel route only if its matched blend adds something useful. Once processing changes the waveform, exact cancellation is no longer a valid quality criterion. Judge the attack, sustain, and musical role of the combination.