न
NAAD Eval
Speech transcript intelligence

Single-file bakeoff

Compare transcripts against ground truth

Upload GT, ElevenLabs, and Naad to measure Strict/Speech WER, surface Naad-only gaps, and see exactly what to improve.

Upload Transcripts

For ASR bakeoff: paste your script text as Ground Truth, ElevenLabs as A, Naad as B. GT is optional for similarity-only mode.

GT Ground Truth (script / reference) (optional)

Click or drag & drop .json / .txt

Optional · needed for WER & gap analysis

A ElevenLabs

Click or drag & drop .json / .txt

Required · baseline

B Naad

Click or drag & drop .json / .txt

Required · our model

C Transcript C (optional)

Click or drag & drop .json / .txt

Optional

Select Transcript Language

Language controls normalization, script handling, numbers, and transliteration.

JSON Key Mapping

Each transcript is analyzed independently. Select the Text key (required) and Speaker key (optional).

Comparison Options

Strict mode is always computed. Configure Non-Strict (Speech) options and speaker mode below.

Comparison Mode

Strict Mode (fixed)

  • Exact word matching
  • Script sensitive
  • Language sensitive
  • Capitalization ignored
  • Special characters removed
  • Number & space normalization
  • No script conversion
  • No transliteration
  • No phonetic matching

Non-Strict (Speech) Options

Phonetic matching is intentionally not available. Code-mix is for Speech WER only — Strict stays script-exact.

Evaluation Dashboard

Selected Language: —

Strict Comparison Metrics

Strict Difference Viewer

Non-Strict Comparison Metrics

Non-Strict Difference Viewer

Calculation Details

Evaluating transcripts...