Maximal Figure-of-Merit Framework to Detect Multi-label Phonetic Features for Spoken Language Recognition