examples on alg.cubing.net?
F2 R U2 R2 B2 L' D2 R U2 R2 U' L' D F' R' D U' B' U'
y z // inspection
B r' U F' E' R' U' F2 // first block (8)
U' R' U R2 U r U R' U' R2 U R U2 r' U' r // second block (16)
R' U' R' F R F' R U' R' U2 R // CMLL (11)
U2 M U M U M' U M' U' M U2 M' U M2 U2 M U2 M // LSE (18)
53 STM
Edit: critiques for the LSE?
I'm not good at Roux, but one thing that you should try to notice is easy LSE EO cases that come up before LSE. In this solve, for example, after your second block, you are left with a 3-move EO (which is much better than the awkward EO case that you actually ended up getting).
So you have two better alternatives:
1) do EO right now, before you do an EO-safe CMLL alg (like COLL for example)
2) do a COLL alg right now, followed by a 3-move EO later on
Pick whichever of the two options would allow you to cancel a move with either
1) the end of the second block
2) the COLL
or 3) the EP step
so for example:
F2 R U2 R2 B2 L' D2 R U2 R2 U' L' D F' R' D U' B' U'
y z //
YOUR inspection
B r' U F' E' R' U' F2 //
YOUR first block (8)
U' R' U R2 U r U R' U' R2 U R U2 r' U' r //
YOUR second block (16)
U L B' U' L2 U' B U B' L2 U B l' U M'//
MY COLL and EO (15)
U D L2 D' M2 D L2 D' //
MY edges (8)
so, that's an easy way to shave off 6 stm, or 13 htm
But Roux is NOT an inherently efficient method. Stock Roux method is highly restrictive, the order of steps doesn't lend itself to half-decent cancellations, and the moveset is gross, to put it gently.