zszyellow / wer-in-python Goto Github PK

Would you be willing to license this under MIT license? Its a great simple tool for wer and I'd like to contribute.
https://choosealicense.com/licenses/mit/#

thanks for your script!
I used to test Chinese, the result was right, but the marked position of EVA was misplaced.

REF: 沉 吟 放 拨 插 弦 中 整 顿 衣 裳 起 敛 容 的 
HYP: 沈 营 放 拨 插 全 中 整 顿 衣 裳 起 脸 容 得 
EVA: S S       S             S   S      
WER: 27.78%

I am trying to calculate a wer for a long text( around 2000 words), I run it on a linux server that's memory is 32G, but it does't return the result. instead of throwing a memory issue. Do you have any way that can improve it ?

The display of aligned text is very useful.

There is no issue with the WER computation. However, I encountered an error message when the Reference.txt word count is more than double of that Hypothesis.txt. A fix for this error will be highly appreciated.

https://github.com/zszyellow/WER-in-python/blob/master/wer.py#L35

My reference has 430 words and the hypothesis has 498 words.
I have already figured out the following:
In the getStepList method, x and y never reach 0 at the same time. If x = 0 then y ~ 250 and if y = 0 then x is way below 0.
Then an infinite number of "d" are appended until the memory overflows but there is no error massage.

How can I fix this problem?

It would be good to mention in README that the library supports computing WER for sequences up to 256 words. Otherwise, the users need to make the change accordingly.

Traceback (most recent call last): File "WER-in-python/wer.py", line 203, in <module> wer(r, h) File "WER-in-python/wer.py", line 191, in wer list = getStepList(r, h, d) File "WER-in-python/wer.py", line 49, in getStepList elif d[x][y] == d[x-1][y-1] and r[x-1] == h[y-1] and x >= 1 and y >= 1: IndexError: index -1054 is out of bounds for axis 0 with size 1053
just tried to run this out of the box and got this