From the README:
Currently, this phase vocoder can create significant phase artifacts when pitch-shifting audio at non-integer factors due to loss of vertical phase coherence. A few alternate algorithms are provided that may be able to reduce this problem but so far I have not found a complete solution.
From the README: