Question about using pyannote/speaker-diarization-3.1 in C++

[zzaahhrraa1994]

Tested versions

I am currently working on a project where I need to use the pyannote/speaker-diarization-3.1 model in C++.

I have tried converting the PyTorch model to ONNX, but I faced some issues and it seems the model cannot be easily converted.

Could you please advise if there is a recommended way to run speaker-diarization in a C++ environment, or any guidance for integrating pyannote models with C++?

Thank you very much for your time and support.

System information

Windows 10-CPU

Issue description

I am trying to export the pyannote/speaker-diarization-3.1 PyTorch model to ONNX format so that I can run segmentation in a C++/ONNX environment. However, the export fails and I cannot generate ONNX-compatible segmentation output.

Expected behavior:
I expect to be able to export the model to ONNX, and run it with ONNX Runtime to obtain the same speech segments (start/end times) as the original PyTorch model, with identical speaker assignments.

Current behavior:

ONNX export fails or produces outputs with no detected segments.

When running ONNX inference, the speech segmentation outputs are empty (Segments detected: 0)

Minimal reproduction example (MRE)

.