Connectionist Architectures for Multi-Speaker Phoneme Recognition