The Belt method has been 'invented' many times before, and it is generally accepted to be a bad method. Here is my attempt at a belt method that is better than the normal belt method.

The first step is, obviously, to create a belt along the E slice. This usually takes around 6 moves.

The second step is to orient all the corners in one algorithm. There are around 150 cases for this step, and it averages about 11 moves, including U and D setup moves.

The third step is to position all the corners in one algorithm. There are around 200 cases for this step, and it averages about 12 moves, including U and D setup moves. Many of the algorithms use only R2, U, and D moves, especially the shorter ones. In fact, five cases are only three moves long! After completing this step, there will only be 8 edges left to solve.

The fourth step is to position the DL and DR edges intuitively. This averages about 5 moves. After this step, the cube can be solved with only M and U moves.

The final step is to solve the last 6 edges like in Roux. This averages about 15 moves.

Total: About 350 algorithms and 49 moves on average. Recognition for the algorithmic steps is pretty bad, so I don't think this method would be viable. At least it's better than the normal belt method.

Example solve: U' L2 R2 D R2 B2 D U2 R2 U2 L' F' L2 R' U F' D U2 B' L2 F' U'
F' U D L' U R' D F' // Belt (8/8)
U R U' D R' U2 D R U' D2 R' // CO (11/19)
D R2 U' R2 D' R2 U R2 // CP (8/27)
D M' D' M D' // DLDR (5/32)
U' M U' M' U' M U' M U2 M' U M' U2 M2 U M2 U2 // LSE (17/49)