Speech synthesis for dysarthric people using a small amount of data and improvement of clarity using a healthy person model