The Sound of My Voice: Speaker Representation Loss for Target Voice Separation