VoxbenchSPEECH RECOGNITION, COMPARED

Find the model that gets you.

One recording. Choose from eleven models. Compare what they hear.

11 contenders
Your accent is the test.
HOW DO YOU WANT TO COMPARE?

Your API keys selected. Comparisons are billed to your provider accounts.

Approve access to your OpenRouter balance. No key to copy.

01 — THE RECORDING

Start with your voice

Up to 60 seconds

Use your everyday pace, names, and technical words. Nothing is sent while recording.

Vocabulary & language0 hints
0/100
Saves automatically as you type. One term per line. Hint support varies by connection.
02 — THE LINEUP

Who’s listening?

8 selected0 with keys entered

Compare sends audio to the selected providers with keys. Each request is billed by that provider. Missing keys are shown as skipped.

03 — TRANSCRIPTS

Your transcripts will appear here.

Reference transcript Optional · what you actually said

Word errors count substitutions, missing words and extra words. Case and punctuation are ignored. Lowercase is a local preview; scores always use the raw transcript.

Record a take, choose your models, then compare.
Real results only — no sample transcripts.