try to figure out how ZeroGPU works

InferenceInterfaces/ToucanTTSInterface.py

@@ -161,11 +161,11 @@ class ToucanTTSInterface(torch.nn.Module):
             phones = self.text2phone.string_to_tensor(text, input_phonemes=input_is_phones).to(torch.device(self.device))
             mel, durations, pitch, energy = self.phone2mel(phones,
                                                            return_duration_pitch_energy=True,
-                                                           utterance_embedding=self.default_utterance_embedding,
+                                                           utterance_embedding=self.default_utterance_embedding.to(device),
                                                            durations=durations,
                                                            pitch=pitch,
                                                            energy=energy,
-                                                           lang_id=self.lang_id,
+                                                           lang_id=self.lang_id.to(device),
                                                            duration_scaling_factor=duration_scaling_factor,
                                                            pitch_variance_scale=pitch_variance_scale,
                                                            energy_variance_scale=energy_variance_scale,