Speaker Extraction on Three-speaker dataset (synthetic)
2.38PESQOracle
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OracleBackbone=SpatialNet, Tracking=true, Params [M]=1.74, MACs [G/s]=18.82026.01 | 2.38 | 84.7 | |
| OracleBackbone=SpatialNet, Tracking=false, Params [M]=1.74, MACs [G/s]=18.72026.01 | 2.27 | 83 | |
| SELDnet with Joint AR (AR-SSF + AR-TST)Backbone=SpatialNet, Tracking=true, AR-SSF=true, AR-TST=true, Params [M]=2.03, MACs [G/s]=18.82026.01 | 2.27 | 82.7 | |
| OracleBackbone=McNet, Tracking=true, Params [M]=1.86, MACs [G/s]=30.32026.01 | 2.21 | 80.4 | |
| adaptive rotary steeringBackbone=SpatialNet, Tracking=true, Params [M]=1.74, MACs [G/s]=18.82026.01 | 2.21 | 82.2 | |
| SELDnet with AR-SSFBackbone=SpatialNet, Tracking=false, AR-SSF=true, Params [M]=2.03, MACs [G/s]=18.82026.01 | 2.18 | 81.6 | |
| SELDnet with Joint AR (AR-SSF + AR-TST)Backbone=McNet, Tracking=true, AR-SSF=true, AR-TST=true, Params [M]=2.15, MACs [G/s]=30.32026.01 | 2.17 | 79.5 | |
| Fixed rotary steeringBackbone=SpatialNet, Tracking=false, Params [M]=1.74, MACs [G/s]=18.72026.01 | 2.11 | 80.3 | |
| SELDnetBackbone=SpatialNet, Tracking=false, AR-SSF=false, Params [M]=2.03, MACs [G/s]=18.82026.01 | 2.11 | 80.6 | |
| OracleBackbone=McNet, Tracking=false, Params [M]=1.84, MACs [G/s]=29.92026.01 | 2.08 | 77.7 | |
| adaptive rotary steeringBackbone=McNet, Tracking=true, Params [M]=1.86, MACs [G/s]=30.32026.01 | 2.08 | 78.5 | |
| SELDnet with AR-SSFBackbone=McNet, Tracking=false, AR-SSF=true, Params [M]=2.13, MACs [G/s]=302026.01 | 2.03 | 76.6 | |
| SELDnetBackbone=McNet, Tracking=false, AR-SSF=false, Params [M]=2.13, MACs [G/s]=302026.01 | 1.98 | 75.5 | |
| Fixed rotary steeringBackbone=McNet, Tracking=false, Params [M]=1.84, MACs [G/s]=29.92026.01 | 1.97 | 75.1 | |
| UnprocessedPipeline=None2026.01 | 1.08 | 31.6 |