Multi-dimensional frequency dynamic convolution with confident mean teacher for sound event detection.