wjn199938 / asr_evaluation_cn

评测asr的识别结果，计算WER和SER

ASR_Evaluation_CN

中文ASR评测指标，WER和SER计算。

Calculate asr evaluation like WER and SER in Chinese. Calculate levenshtein distance.

Requirements

• Python 3.7+

WER

WER: Word Error Rate

$$ WER=\frac {S+D+I}{N} = \frac {S+D+I}{S+D+C} $$

• S: the number of subsitutions. 替换的数量
• D: the number of deletions. 删掉的数量
• I: the number of insertions. 插入的数量
• C: the number of correct words. 正确的数量
• N: the number of words in the reference. 总字数

SER (Same like flow)

SER: Sentence Error Rate

$$SER=\frac {S+D+I}{N} = \frac {S+D+I}{S+D+C}$$

• S: the number of subsitutions.
• D: the number of deletions.
• I: the number of insertions.
• C: the number of correct words.
• N: the number of words in the reference.

Usage

git clone [email protected]:yanhuibin315/ASR_Evaluation_CN.git
python demo.py

例子

标准结果：

["你好", "**"]

识别结果：

["你嗨", "**"]

WER = 1/4 = 0.25

SER = 1/2 = 0.5