WordErrorRate

Models & Types · AiDotNet.Metrics

Word Error Rate (WER) metric for speech recognition evaluation.

How It Works

WER measures the accuracy of automatic speech recognition (ASR) systems by computing the edit distance between the predicted transcription and the reference transcription.

Formula: WER = (Substitutions + Insertions + Deletions) / Number of words in reference

Typical WER values:

  • <5%: Excellent (human-level performance)
  • 5-10%: Very good
  • 10-20%: Good
  • 20-30%: Acceptable
  • 30%: Poor

Constructors

Constructor Summary
WordErrorRate()

Methods

Method Summary
ValueTuple<double, int, int, int, int> ComputeDetailed(string hypothesis, string reference) Computes detailed error statistics including substitutions, insertions, and deletions.
double Compute(string hypothesis, string reference) Computes the Word Error Rate between a hypothesis and reference transcription.
double ComputeBatch(string[] hypotheses, string[] references) Computes WER for a batch of hypothesis-reference pairs.