We measured PP-P differential travel times on broadband seismograms of the Incorporated Research Institutions for Seismology (IRIS) network. The measurement was made by cross-correlating the observed PP waveform with the synthetic PP calculated from the observed P waveform. The effect of seawater reverberation on the measurement is taken into account if the PP bounce point is under ocean. The PP-P times measured after low-pass filtering is systematically shorter than those measured on raw broadband seismograms. We show that low-pass filtering tends to merge the precursory and postcursory phases into the main PP phase so that the measured PP-P times are artificially short. A cutoff frequency as high as 0.5 Hz is required to isolate the main PP. With such a high cutoff frequency, we obtained a total of about 7000 PP-P times. The residuals relative to a reference P velocity model are consistent with those calculated for the aspherical model WEPP2 <Obayashi et al., 1997> when the PP bounce points are located in the region well resolved by WEPP2. The PP-P times, combined with 2 million International Seismological Centre (ISC) first arrival times, were inverted for new model WEPP2'. This model renders a variance reduction of PP-P residuals of 63% without compromising the fit to P residuals used originally in WEPP2. In WEPP2', the low velocity anomaly beneath Hawaii is limited in depth down to the uppermost part of the lower mantle. The high velocity anomaly of the Russian continental lithosphere is underlain in the mantle transition zone by a body of low velocity anomaly.