Local Feature Extractors Accelerating HNNP for Phoneme Recognition

摘要: Artificial neural networks are fast in the application phase but very slow training phase. On other hand there state-of-the-art approaches using networks, which efficient image classification tasks, like hybrid network plait (HNNP) approach for images from signal data stemming instance phonemes. We propose to accelerate HNNP phoneme recognition by substituting with highest computation costs, convolutional network, within a preceding local feature extractor and simpler faster network. Hence, this paper we appropriate extractors problem investigate compare resulting costs as well performance. The results of our experiments show that best one proposed combination smaller is more than two times complex delivers still good