Binbin Zhang

Just use conf/train_conformer.yaml for your experiment. please see https://github.com/wenet-e2e/wenet/blob/main/examples/aishell/s0/run.sh for demo.  And the only configure difference of none-streaming model and streaming model is the "use_dynamic_chunk" option in configure yaml.

The result is weird, I think maybe the result are both from non-streaming conformer tested with chunk 16. In theory, you can't test the non-streaming model with fixed chunk, the...

It's a reference implementation now, so we did less work on checking the invalid protocal and error processing. Could you help on this based on your case?

how many epochs now? maybe: 1. the sudden jump is caused by some bad data. 2. the training is not stable enough, try to use smaller learning rate. We have...

One possible reason is that you did not set finish signal when your VAD system triggers. please see https://github.com/wenet-e2e/wenet/blob/main/runtime/core/websocket/websocket_server.cc#L59

> @robin1001 Are you sure that the current code supports flushing ASR outputs when signaling "end" to the websocket with chunk_size > 0? > > https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/torch_asr_decoder.cc#L106 Here we set state...

Is it the latest code? It should be close by https://github.com/wenet-e2e/wenet/blob/main/wenet/dataset/processor.py#L109.

Please help to make a PR, thanks

@hicliff

Have you updated the code to the latest version? There is a bug before, and we just fix it, please try the latest code.