Target speaker anonymization in multi-speaker recordings