Google Summer of Coding Week 4&5

This is my 4th blog in this category (Google Summer of Code 2018). I am writing the reports for week 4 and week 5 together this time. Since finals weeks are coming, so I am busy with tests these days. But still, I was able to follow up the planned schedule. Let us start. 🙂

First Evaluations and Updated Chinese Pipeline

Just after the 4th week, it was the time for the first GSoC evaluations. Therefore, week 4 is mainly about document our progress on Chinese pipeline.

I establish a Github repository for GSoC other than this blog. I’ve included our work these days in the project’s README. If you like, you can have a look.

DeepSpeech2 Paper Review

Our Chinese language model based on Baidu’s Deep Speech 2 paper, with PaddlePaddle platform. Since we will directly use the released Mandarin LM Large as our language model, it is important for us to fully understand its operation principle.

The author list of this paper is really long, and this paper contains large amount of work, from architecture design to product deployment. This paper details their contribution to the model architecture, large labeled training datasets, and computational scale for speech recognition. This includes an extensive investigation of model architectures, and our data capturing pipeline that has enabled us to create larger datasets than what is typically used to train speech recognition systems.

Running in screen on gallina

When I tried to run DeepSpeech2 on CWRU server, I ran into the problem like:

packet_write_wait: Connection to 10.57.19.250 port 22: Broken pipe

A useful utility which can be used to mitigate problems caused by this is called screen. Screen is a powerful utility that allows you to control multiple terminals which will stay alive independently of the ssh session. Actually, Prof. Turner taught me to use screen before, but I didn’t pay much attention at that time (feel so guity about that!).

Here’s the screen command which might be useful as reference:

1. Login ssh

CynthiaHsus-MacBook-Pro:~ suwi$ ssh sxx186@redhen1.case.edu
Last login: Thu Jun 28 03:47:14 2018 from 122.245.123.171
sxx186@redhen1:~$ ssh sxx186@rider.case.edu
Last login: Thu Jun 28 03:42:44 2018 from redhen1.cosi.cwru.edu

2. Enter screen

[sxx186@hpclogin ~]$ screen

3. Exit screen

control a + d

4. Recover screen

screen -r

5.Other useful command

see screen information

screen -ls

kill a screen dialogue

screen -X -S [session # you want to kill] quit

create new screen

control a + c

switch to the next screen

control a + n

Conclusion

This wraps up my report for the 4th and 5th week. I will be writing about the 6th and 7th week soon. Thank you for reading. 😀