Skip to main content
Building with speech-to-text doesn’t stop at transcription — you need to understand how well your system performs in real-world conditions. To help you, we want to share some resources that will help you identify best practices for evaluating speech to text (and we built a tool to help you do it too!). Evaluation helps you: Measure accuracy beyond simple metrics like WER
Identify failures on critical terms (e.g., medical vocabulary)
Compare models, configurations, and prompts reliably
Debug systematic errors and improve product quality

​
A Better Way to Evaluate ASR

Metrics like Word Error Rate (WER) treat all words equally — but in real-world use cases, not all errors carry the same weight. Missing filler words is very different from missing a diagnosis, medication, or key symptom. ErrorAlign is a next-generation alignment method designed for modern speech recognition. Instead of forcing transcripts into rigid, one-to-one comparisons, it produces more natural, human-like mappings between reference and model output — making error analysis clearer, more reliable, and more actionable.

Click here to learn about Error Align

    Produces more human-like transcript comparisons
    Improves error attribution
    Enables deeper, more actionable analysis

​
Use Corti Canal to Analyze for you

To make this practical, we’ve open-sourced Corti Canal, a tool for evaluating ASR outputs using Error Align. Don’t spend your time on figuring out how to evaluate and jump right into evaluation.
Corti Canal

Click here to Install Corti Canal

    Built-in normalization
    Advanced alignment methods
    Visual error analysis
    Domain-specific metrics (e.g., medical term recall)

​
What This Enables

Instead of asking:
“What’s my WER?”
You can answer:
“Where is my model failing — and does it matter?”
For best results, evaluate using normalized text, advanced alignment, and domain-specific metrics.
Happy testing!