Skip to main content
Lightning Jar - Web Studio Lightning Jar Wordmark

Findings No. 84

L/R as two pseudo-stems is not worth it

L and R channels correlate too highly to work as separate pseudo-stems in Stem Shovel; mono beat the two-channel split in study S33-G2.

Stem Shovel Browser Platform verified

Treating the left and right channels of a stereo mix as two separate pseudo-stems does not pay for itself. The idea was to feed transcription the L and R channels as two inputs instead of one mono mix, on the theory that the channels might isolate different parts of the band.

In study S33-G2 (PASS), the two-channel approach scored 73/76, one bar worse than plain mono, while costing twice the transcription work since each channel runs its own pass. The corpus explains why: L and R correlate at 0.83 to 0.90, the side channel sits 10 to 13 dB down from the mid, and the low end is centered. Each channel is a near-copy of the whole band, not a separate part of it, so splitting them duplicates signal and cost without adding information.

The fix was to keep things simple: Stem Shovel runs transcription on the mono mix rather than on L and R as separate pseudo-stems.

The general rule: before trading one input for two, check whether the two actually disagree. If they correlate above about 0.8, you are paying for a duplicate, not a new signal.

Receipt: Stem Shovel, study S33-G2 (stem-shovel).

Ask Eljay

Ask about the studio's work, research, packages, or writing. A few starters:

Answers come from this site's own content and link their sources. For anything that matters, email hello@lightningjar.com.