Fangjun Kuang

Results 683 comments of Fangjun Kuang

> I met the other error when using the "make" command . c++: error: ../online2/libkaldi-online2.so: No such file or directory : recipe for target 'nnet3-init' failed make[1]: *** [nnet3-init] Error...

> I am not getting (kaldi.mk line 46) from https://github.com/alphacep/kaldi/blob/master/src/Makefile > Not getting exact line, what should I remove, can you make me clear. Before you run ``` root@kush-Linux-1:/opt/kaldi/src# make...

I think you can get more help if you post all the logs of `make`.

Can you use ``` make -j clean make -j1 ``` and post the log of `make -j1` ?

Instead of doing ```cpp std::vector d_arc_nextstates_; ``` How about wrapping it inside a custom class, e.g., ```cpp template class Foo { public: // we can control which methods should be...

I guess the above comments miss the point of the motivation why @kkm000 wants to do this. What he wants is to avoid memory leaks, not the usages of`operator()[]`, `at()`,...

> @danpovey, is pybind11 still a thing? Not sure. With the experience gained from k2, I think we can do a better job about wrapping kaldi to Python (e.g, integration...

> > Not sure. With the experience gained from k2, I think we can do a better job about wrapping kaldi to Python (e.g, integration with PyTorch). > > Rustlang...

I also find there is a post at https://speechbrain.discourse.group/t/expected-speed-on-cpu-and-gpu/113 asking the information about the decoding speed using `asr-crdnn-rnnlm-librispeech`. However, collaborators of this repo don't give any statistics about the actual...

By the way, https://huggingface.co/speechbrain/asr-transformer-transformerlm-librispeech says the WERs for `test-clean` and `test-other` are I downloaded the model and the `hyperparams.yaml` config file without changes but get a worse WER ``` 2021-08-12...