WER

Introduction

This repo contains the scripts to calculate best Word Error Rate(WER). WER is an indicator of the accuracy for the automatic speech recognition(ASR) system.

Background

The Word Error Rate (WER) is derived from the Levenshtein distance. It calculates how far the hypothesis differs from the reference, which shows the accuracy of ASR. Each the hypothesis word sequence is aligned with the reference word sequence, and WER is computed based on the number of substitutions, deletions and insertions needed.