Fangjun Kuang

> Thanks for the heads up about the torchaudio loss! @snakers4 You may find https://github.com/danpovey/fast_rnnt useful.

I did not mean that you are using the name `TCN` and `max-pooling` loss. What I mean is that the network architecture and parameters, i.e., the values of - number...

Can you try https://github.com/csukuangfj/optimized_transducer ? It uses less memory and may fix your issue.

There is also an implementation at https://github.com/csukuangfj/optimized_transducer that uses less GPU memory.

Please have a try with https://github.com/csukuangfj/optimized_transducer It has a consistent gradient for CPU and CUDA.

@npuichigo We have an ongoing PR #99 , almost finished, about C++ APIs for non-streaming ASR. The streaming part is still under development. You can find an example usage by...

@npuichigo Do you have an interest in using https://github.com/chriskohlhoff/asio with c++20 coroutine? It may take years to get the above proposal accepted by the c++ standard.

> It appears that the server gets stuck on long files such as 10 minutes of Audio. Are you using the offline server or the streaming server? Are you using...

I think the streaming version should be able to handle very long utterances. I will test it locally to see whether it is reproducible.

> It appears that the streaming server expects to know the length of audio upfront. What do you mean by this? What the server needs to know is the number...