Decoding with Hotwords giving segmentation fault

[chiragpatel39]

29166 Segmentation fault python python-api-examples/online-decode-files.py --encoder $model_dir/encoder-epoch-19-avg-1.onnx --decoder $model_dir/decoder-epoch-19-avg-1.onnx --joiner $model_dir/joiner-epoch-19-avg-1.onnx --tokens $model_dir/tokens.txt --num-threads=1 --decoding-method=modified_beam_search --hotwords-file=local/hotwords.txt --modeling-unit bpe ${wav}"003".wav ${wav}"004".wav ${wav}"005".wav ${wav}"006".wav ${wav}"007".wav ${wav}"008".wav ${wav}"009".wav ${wav}"010".wav

My hotword file looks like:

VMO :2.0
quarantine :2.0
remdesivir :2.0
oseltamivir :2.0
isolation :2.0
n ninety five :2.0
NRBM :2.0

Regards,
Chirag Patel

[csukuangfj]

What is model_dir?

[csukuangfj]

Also, are you able to reproduce the hotwords example in the documentation?

[csukuangfj]

Also, please tell us how to reproduce your issue. (You need to provide more details)

[chiragpatel39]

model_dir is the directory which contain checkpints of trained model.
No I haven't tried with the model given in example. I am following example with our own model.
I will try the example with pre-trained model.

[csukuangfj]

If you want to debug it with your own model, please try to use C++ binaries. You can use gdb to debug it.

[vsd-vector]

I think this could related to my last merged PR (preserving context between endpointing).

If it is, I will try to create fix as soon as possible.

[csukuangfj]

I think this could related to my last merged PR (preserving context between endpointing).

If it is, I will try to create fix as soon as possible.

Thanks!