We develop a computation reduction method for the real root method that is mainly used in the CELP (code excited linear prediction) vocoder. The real root method is that if polynomial equations have real roots, we are able to find those and transform them into LSP (line spectrum pairs). However, this method takes much time to compute, because the root searching is processed sequentially in the frequency region. But the important characteristic of LSP is that most coefficients occur in a specific frequency region. We suggest a method for reducing the LSP transformation time using voice characteristics. In the case of the silence period, the distribution of LSP parameters is uniform. In the case of the voice activity period, distribution of LSP parameters is not sequential. The proposed method applies search order and interval differently according to the distribution of LSP parameters. Also in the case of vowels, the search band can be controlled by the relation between the first formant and the second formant. When we compare this proposed method with the conventional real root method, the experimental result is that the searching time was reduced about 46% on average.