Efficient compression and reconstruction of speech signals using compressed sensing