Discriminate natural versus loudspeaker emitted speech