Improving the kelly-lochbaum vocal tract model using conical tube sections and fractional delay filtering techniques.