problem of synthesise speech with Python and GPU card

[jinsongpan]

when i use the "./src/test_lpcnet.py" to synthesise speech with GPU, the synthesis speed is extremely slow. after serveral hours, i interrupt the program and found it stops at "p, state1, state2 = dec.predict([fexc, iexc, cfeat[:, fr:fr+1, :], state1, state2])". what's wrong with the script? can anyone execute the script successfully?
what's more, when i synthesise speech with c and cpu script "test_lpcnet", it run very well and fast.