I–vector transformation and scaling for PLDA based speaker recognition