VoxBench
voxbench.karnavpargi.com
I built VoxBench to replace guesswork when choosing a speech-to-text model. A team can record one answer, run the same audio through every available model, and compare the results under identical conditions.
It turns a fragmented model-selection exercise into one repeatable benchmark, helping engineering teams weigh transcript quality against cost and response time before committing to a provider.
- Side-by-side OpenAI and Google model evaluation
- Per-model transcript, rating, token, cost, and latency results
- Bring-your-own-key flow with credentials kept in the browser
- Past-run history for consistent model comparisons




