Compare event_error_rate and word_error_rate

[ChristianGeng]

Problem

event_error_rate and word_error_rate can have identical results:

ref = [["hello", "world", "abc"]]
ans = [["hello", "moon"]]

print("eer", audmetric.event_error_rate(ref, ans))
print("wer", audmetric.word_error_rate(ref, and))

results in

eer 0.6666666666666666
wer 0.6666666666666666

When do these two implementations differ, and should this be documented?

Originally posted by @ChristianGeng in #63 (comment)

[ChristianGeng]

git blame tells me that this was implemented this in summer 2021.
As soon as #63 will have been merged, the issue will persist.

ref = [["hello", "world", "abc"]]
ans = [["hello", "moon"]]

print("eer", audmetric.event_error_rate(ref, ans))
print("wer", audmetric.word_error_rate(ref, ans))
print("eer", audmetric.word_error_rate(ref, ans, norm="longest"))
print("eer", audmetric.word_error_rate(ref, ans, norm="truth"))

looks good

eer 0.6666666666666666
wer 0.6666666666666666
eer 0.6666666666666666
eer 0.6666666666666666

However when trying to use word_eror_rate as a replacement for event_error_rate in its tests:

def test_event_error_rate(truth, prediction, eer):
    np.testing.assert_equal(audmetric.word_error_rate(truth, prediction, norm="longest|truth"), eer)

I get some non passing tests.

So it seems as it is not as trivial as replacing the event_error_rate by word_error_rate and deprecating it.