February 9, 2026

CER Metrics and RTFx Performance Tracking Added

Character Error Rate and Real-Time Factor metrics are now available alongside WER in a comprehensive Results Summary table.

What's new

We've expanded our quality metrics with Character Error Rate (CER) calculation and added Real-Time Factor (RTFx) performance tracking. Both are now part of the unified Results Summary table on the /wer page.

This gives you a complete picture of STT provider performance: accuracy metrics (CER and WER) alongside speed metrics (RTFx).

New metrics explained

Character Error Rate (CER)

CER measures errors at the character level, making it especially useful for:

Languages without clear word boundaries (Chinese, Japanese, Thai)
Assessing punctuation and formatting accuracy
Fine-grained quality analysis beyond word-level metrics

Like WER, CER is available in Raw and Normalized modes to match your evaluation needs.

Real-Time Factor (RTFx)

RTFx shows how fast transcription happens relative to audio duration:

RTFx < 0.5× — Very fast (processes faster than 2× audio speed)
RTFx < 1.0× — Real-time (keeps up with audio playback)
RTFx < 2.0× — Slower than real-time but reasonable
RTFx ≥ 2.0× — Slow (takes more than 2× audio duration)

RTFx helps you balance accuracy and speed when choosing STT providers for your use case.

Results Summary table

The new Results Summary table displays all metrics in one place:

CER Raw — Character-level accuracy with exact comparison
CER Normalized — Character-level accuracy with text normalization
WER Normalized — Word-level accuracy with Nvidia NeMo normalization
WER Strict — Word-level accuracy with strict normalization rules
Process time — How long transcription took (seconds)
RTFx — Real-Time Factor with color-coded performance indicators

Lower error rates (CER/WER) mean better accuracy. Lower RTFx means faster processing.

Try it now

Head to /wer, upload your audio with a reference transcript, and compare providers across all quality and performance metrics.