Unsupervised intra-speaker variability compensation based on Gestalt and model adaptation in speaker verification with telephone speech