About Voxa

Voxa gives feedback on how you sound when you speak: for a talk, a job interview, a presentation or a recorded video. You record yourself or drop in a file, and it returns eight scores on your delivery (confidence, charisma, authority, clarity, warmth, enthusiasm, professionalism and nervousness), each with the measurements that produced it.

It is free and needs no account. The analysis runs on your device, so your recording is never uploaded.

Voxa keeps two kinds of numbers apart. Measurements such as pitch, loudness, pauses, speaking rate, jitter and shimmer are computed directly from the sound wave and are reproducible. Scores such as confidence or charisma cannot be measured directly: they are inferred from those measurements by documented rules.

A score is therefore a read of what your delivery signals and why, not a verdict about you. Accent, language, microphone and room all move the numbers, and the bands are set for English speech. The most reliable use is comparing your own recordings made the same way. Every rule is written out on the methodology page.

Voxa is made by Mathieu Dutour Sikirić, who runs MSM programming. Mathieu is a mathematician trained at the École Normale Supérieure in Paris, with a PhD from Université Paris-Sud, and the author of over a hundred research papers and four books. He spent twelve years as a research scientist, and since 2018 has worked as a software engineer on compilers, trading systems, blockchain protocols and optimisation.

That background shapes the tool: each score is a transparent formula over quantities you can check, and the methodology states what the numbers cannot tell you.

The measurements and scores are computed by a Rust library compiled to WebAssembly, so the same code runs in every modern browser. The optional transcription, which adds speaking rate and filler words, uses OpenAI’s open Whisper speech model, also running inside the page. The site is static: there is no server-side processing, no database and no user account.

Voxa is free to use and is supported by advertising from Google AdSense. Ads never change the analysis, and your recording is never uploaded. The privacy policy explains how the ads use cookies and how to control them.

Questions, bug reports, suggestions and corrections are welcome at [email protected]. If a score looks wrong, describe the recording (length, language, microphone) rather than sending it: we never need your audio.

MSM programing d.o.o.
Karlovačka cesta 28B, 10452 Klinča Sela, Croatia