PAPER Special Section on Advances in Modeling for Real-world Speech Information Processing and its Application Structured Adaptive Regularization of Weight Vectors for a Robust Grapheme-to-Phoneme Conversion Model