Automatic Extraction of Geometric Lip Features with Application to Multi-Modal Speaker Identification