wavecap-evaluate

To calculate Word Error Rate (WER) for audio transcription accuracy, you compare generated text against a ground truth transcript. This evaluates output quality by also computing Character Error Rate and analyzing segment-level probabilities.

Yes, you can run regression testing on audio transcriptions by exporting audio files alongside expected transcripts. This enables running automated tests against known datasets to compare metrics like WER before and after system updates.

To evaluate transcription model regression, you should use Word Error Rate, Character Error Rate, and transcription rates. Analyzing confidence scores and segment-level probabilities provides deeper insight into specific areas of degradation.

To verify a Whisper model update has not degraded transcription accuracy, run a suite of regression tests against a known dataset and compare the Word Error Rate before and after the update to ensure reliable output.

Yes, audio analysis includes segment-level probability and confidence score evaluation. This feature breaks down transcription results to assess reliability and pinpoint exact segments where accuracy may be compromised.