Calculates metrics for Speech-To-Text systems. Available metrics: WER, WCR, SER.
Inspired by @imalic3/python-word-error-rate project.
$ npm install @mcvlad/speech-metrics
const calculateMetrics = require("@mcvlad/speech-metrics");
// Each pair consists of: "reference" (input string) and "hypothesis" (output string)
const inputData = [
[`Indigenous Australians have eaten kangaroos for millenia.`, `Indigenous australian sevington kangaroos for millennia`],
[`Welcome to Australia.`,`Welcome to australia`]
];
const metrics = calculateMetrics(inputData);
console.log(metrics);
/* =>
{
results: [{
hypothesis: "indigenous australian sevington kangaroos for millennia",
reference: "indigenous australians have eaten kangaroos for millenia",
WCR: 0,
WER: 0.5714285714285714
}, {
hypothesis: "welcome to australia",
reference: "welcome to australia",
WCR: 1,
WER: 0
}],
SER: 0.5555555555555556,
WER: 0.22857142857142854,
WCR: 0.3333333333333335
}
*/WER detects the percent of recognition errors.
WCR detects the percent of correctly recognized words and ignores Insertion errors.
SER shows the percent of sentences with 1 or more errors.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent