Spaces:
Running
Running
File size: 5,159 Bytes
eaeed82 53fc5a5 b5095e1 53fc5a5 eaeed82 49b38c8 eaeed82 53fc5a5 b5095e1 7ceaf30 2e8c299 b5095e1 53fc5a5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 |
---
title: CER
emoji: 🤗
colorFrom: blue
colorTo: red
sdk: gradio
sdk_version: 3.19.1
app_file: app.py
pinned: false
tags:
- evaluate
- metric
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).
```python
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](#Examples) below).
### Values from popular papers
This metric is highly dependent on the content and quality of the dataset, and therefore users can expect very different values for the same model but on different datasets.
Multilingual datasets such as [Common Voice](https://huggingface.co/datasets/common_voice) report different CERs depending on the language, ranging from 0.02-0.03 for languages such as French and Italian, to 0.05-0.07 for English (see [here](https://github.com/speechbrain/speechbrain/tree/develop/recipes/CommonVoice/ASR/CTC) for more values).
## Examples
Perfect match between prediction and reference:
```python
from evaluate import load
cer = load("cer")
predictions = ["hello world", "good night moon"]
references = ["hello world", "good night moon"]
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
0.0
```
Partial match between prediction and reference:
```python
from evaluate import load
cer = load("cer")
predictions = ["this is the prediction", "there is an other sample"]
references = ["this is the reference", "there is another one"]
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
0.34146341463414637
```
No match between prediction and reference:
```python
from evaluate import load
cer = load("cer")
predictions = ["hello"]
references = ["gracias"]
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
1.0
```
CER above 1 due to insertion errors:
```python
from evaluate import load
cer = load("cer")
predictions = ["hello world"]
references = ["hello"]
cer_score = cer.compute(predictions=predictions, references=references)
print(cer_score)
1.2
```
## Limitations and bias
CER is useful for comparing different models for tasks such as automatic speech recognition (ASR) and optic character recognition (OCR), especially for multilingual datasets where WER is not suitable given the diversity of languages. However, CER provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort.
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
```bibtex
@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.}
}
```
## Further References
- [Hugging Face Tasks -- Automatic Speech Recognition](https://huggingface.co/tasks/automatic-speech-recognition)
|