Help me fix running time error in a deep learning C++ project

I have a VS2017 C++ project, which converted from runnable python codes. It is supposed to predict the direction of arrival of sound via several microphone audio files, and now I need someone to fix bugs to make the project running successfully and correctly.

As shown in the attached is the the first problem I meet:
mic_signals should be a 15xL tensor, however, now it’s a Lx1 tensor. What’s the reason and how to solve it?

Plus, the definition of mic_signals_a: torch::Tensor mic_signals_a[15];

For following problems, full project codes will be provided.