Speech transients are important cues for identifying and discriminating speech sounds, and several studies have suggested that selective amplification of these transients can improve the intelligibility of speech in noise. This paper describes an improved version of a wavelet-based method for extracting transient speech that we described in  and the use of the articulation index to select optimal parameters for the method. The new method combines subband decomposition by wavelet packets and transition rate characterization based on the first derivative of short-time energy. The method also incorporates a threshold which, when varied, controls the amount of quasi-steady-state activity that is included in the transient speech signal. The speech modification scheme is optimized and intelligibility improvement is estimated using the articulation index.