title: CER
emoji: 🤗🏃🤗🏃🤗🏃🤗🏃🤗
colorFrom: blue
colorTo: red
sdk: gradio
sdk_version: 3.19.1
app_file: app.py
pinned: false
tags:
- evaluate
- metric
license: apache-2.0
description: >-
Character error rate (CER) is a common metric of the performance of an automatic speech recognition system.
CER is similar to Word Error Rate (WER), but operates on character instead of word. Please refer to docs of WER for further information.
Character error rate can be computed as:
CER = (S + D + I) / N = (S + D + I) / (S + D + C)
where
S is the number of substitutions,
D is the number of deletions,
I is the number of insertions,
C is the number of correct characters,
N is the number of characters in the reference (N=S+D+C).
CER's output is not always a number between 0 and 1, in particular when there is a high number of insertions. This value is often associated to the percentage of characters that were incorrectly predicted. The lower the value, the better the
performance of the ASR system with a CER of 0 being a perfect score.
Metric Card for CER
Metric description
Character error rate (CER) is a common metric of the performance of an automatic speech recognition (ASR) system. CER is similar to Word Error Rate (WER), but operates on character instead of word.
Character error rate can be computed as:
CER = (S + D + I) / N = (S + D + I) / (S + D + C)
where
S is the number of substitutions,
D is the number of deletions,
I is the number of insertions,
C is the number of correct characters,
N is the number of characters in the reference (N=S+D+C).
How to use
The metric takes two inputs: references (a list of references for each speech input) and predictions (a list of transcriptions to score).
from evaluate import load
cer = load("cer")
cer_score = cer.compute(predictions=predictions, references=references)
Output values
This metric outputs a float representing the character error rate.
print(cer_score)
0.34146341463414637
The lower the CER value, the better the performance of the ASR system, with a CER of 0 being a perfect score.
However, CER's output is not always a number between 0 and 1, in particular when there is a high number of insertions (see Examples below).
Values from popular papers
Examples
Perfect match between prediction and reference:
!pip install evaluate jiwer
from evaluate import load
cer = load("cer")
predictions = ["hello világ", "jó éjszakát hold"]
references = ["hello világ", "jó éjszakát hold"]
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
0.0
Partial match between prediction and reference:
from evaluate import load
cer = load("cer")
predictions = ["ez a jóslat", "van egy másik minta is"]
references = ["ez a hivatkozás", "van még egy"]
cer = evaluate.load("cer")
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
0.9615384615384616
No match between prediction and reference:
from evaluate import load
cer = load("cer")
predictions = ["üdvözlet"]
references = ["jó!"]
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
1.5
CER above 1 due to insertion errors:
from evaluate import load
cer = load("cer")
predictions = ["Helló Világ"]
references = ["Helló"]
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
1.2
Limitations and bias
.
Also, in some cases, instead of reporting the raw CER, a normalized CER is reported where the number of mistakes is divided by the sum of the number of edit operations (I + S + D) and C (the number of correct characters), which results in CER values that fall within the range of 0–100%.
Citation
@inproceedings{morris2004,
author = {Morris, Andrew and Maier, Viktoria and Green, Phil},
year = {2004},
month = {01},
pages = {},
title = {From WER and RIL to MER and WIL: improved evaluation measures for connected speech recognition.}
}