Wei Kang

I got the same error. （tensorflow1.7, window10(64),python3.5）

@shaynemei Did you use `decode-right-context=2` (the default value) in sherpa. If so, please try `decode-right-context=0`. We found that not all models can benefit from right context.

Also, can you show your decoding command for local batch decoding and local streaming decoding, I think the WER difference between them is a little large. Thanks!

> There are about 0.3 to 0.4 seconds delay for the streaming model. Maybe we can try the models trained with delay penalty, see if it helps.

The following are the timestamps of model trained with delay penalty (see https://github.com/k2-fsa/k2/pull/976). The encoder is reworked conformer. You can see from the table, the model trained without penalty delay,...

~~@shaynemei Try smaller decode-chunk-size? like:~~ ``` --decode-chunk-size=8 \ --left-context=32 \ --right-context=0 ``` [edit:] Sorry, the default chunk-size is 8 already.

This is a ready version for `GreedySearch` decoding method (i.e. without k2 core dependency). We will merge this when we fix the issues of building a standalone k2 core library...

I don't think we need sherpa for android / ios apps, k2 is enough, too many dependancy makes an application complicated. I agree to create a new android-demo project. @csukuangfj...

How do you run it with fast_rnnt? Can you show your running command? Looks like to me, it is the wrong configuration of CUDA devices.

> For data security reasons, i can't show the code here, but only part of it. Thanks！I see. > export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7,8" BTW, do you have 9 cards in your machine?...