The word scores are not to be ignored, though. The p value was 0.01, which means these results were very unlikely to be due to chance.
One possible way to interpret this is that the drug works but doesn't diffuse far enough in the dose provided for the safety study (ie only frequencies higher than 8 kHz were involved except for those 4 who got 10 dB improvements).
Since the speech in noise scores also improved, it may have an even greater effect on synaptic connections than the hair cells. So many unknowns but with the p value provided, it is statistically unlikely the drug has no effect on hearing.