Liyong.Guo

@desh2608 @ngoel17 Could you submit the lattice so I could reproduce your error?

@desh2608 @ngoel17 You could save the lattice with following statement and share it to us with goole drive. ``` torch.save(lattice.as_dict(), "lattice_triggering_shortest_path_error.pt") ```

Thanks for the report! > After this, the decoding could run successfully, but the WER was worse than greedy search. I have two questions about this: Q1: Was the WER...

The erroneous lattice here: [error_lattice.tar.gz](https://github.com/k2-fsa/k2/files/9861926/error_lattice.tar.gz) is non-connected. You could also verify this with following code: ``` import k2 import torch lattice_state = torch.load('lattice_triggering_shortest_path_error.pt') lattice = k2.Fsa.from_dict(lattice_state) connected_lattice = k2.connect(lattice) #...

> I'll package the model and the input corresponding to the error lattice and attach it here in a bit Thanks! It would be much better if you also provide...

> Let me know if you need any more details. Thanks for your package! I could reproduce the error now.

@ncakhoa Differences: 1. The main difference is k2.nbest support autograd, while the one in icefall doesn't. 2. nebst in icefall may have more functions related to decoding 3. k2.nbest is...

@jiaji-huang Could you help us to save a lattice which could reproduce this problem and submit to us? An example lattice saving statement: ``` torch.save(lattice.as_dict(), "lattice_triggering_shortest_path_error.pt") ``` It's easier for...

@jiaji-huang Would mind try to add k2.connect or k2.arc_sort(k2.connect) before k2.shortest_path? An example may looks like this: ``` k2.shortest_path(k2.connect(lattice), use_double_scores=False) or k2.shortest_path(k2.arc_sort(k2.connect(lattice)), use_double_scores=False) ``` Also, I think there might be...

> using beam_size =10 at creating the lattice avoids the segmentation fault What's the original beam_size? > Not sure if I did everything in a right way. Maybe you could...