Word embedding is essential for neural network models for various natural language processing tasks. Since the word embedding usually has a considerable size, in order to deploy a neural network model having it on edge devices, it should be effectively compressed. There was a study for proposing a block-wise low-rank approximation method for word embedding, called GroupReduce. Even if their structure is effective, the properties behind the concept of the block-wise word embedding compression were not sufficiently explored. Motivated by this, we improve GroupReduce in terms of word weighting and structuring. For word weighting, we propose a simple yet effective method inspired by the term frequency-inverse document frequency method and a novel differentiable method. Based on them, we construct a discriminative word embedding compression algorithm. In the experiments, we demonstrate that the proposed algorithm more effectively finds word weights than competitors in most cases. In addition, we show that the proposed algorithm can act like a framework through successful cooperation with quantization.

This paper describes the creation of a dysarthric speech database which has been done as part of a national program to help the disabled lead a better life ― a challenging endeavour that is targeting development of speech technologies for people with articulation disabilities. The additional aims of this database are to study the phonetic characteristics of the different types of the disabled persons, develop the automatic method to assess degrees of disability, and investigate the phonetic features of dysarthric speech. For these purposes, a large database of about 600 mildly or moderately severe dysarthric persons is planned for a total of 4 years (2010.06. 01 ~ 2014.05.31). At present a dysarthric speech database of 120 speakers has been collected and we are continuing to record new speakers with cerebral paralysis of mild and moderate severity. This paper also introduces the prompting items, the assessment of the speech disability severity of the speakers, and other considerations for the creation of a dysarthric speech.