Voice Studio — Speech to Text & Read Aloud
Dictate and get a clean transcript, or paste text and have it read out loud. The speech model runs on your own device, so nothing you say is sent anywhere.
Three quick steps
1 · Allow the mic
Your browser asks once. Audio is captured locally and never transmitted.
2 · Speak
Press the mic and talk. The model transcribes on your device when you stop.
3 · Use it
Copy the transcript, polish the grammar, or send it to text-to-speech.
About Voice Studio
Is my voice sent to a server?
No. Recording uses your browser’s microphone API and transcription runs a Whisper model compiled to run inside the page. The audio never leaves your device.
Why does the first recording take a moment?
The speech model is about 40 MB and downloads once. Your browser caches it afterwards, so later recordings start straight away.
How accurate is it?
It uses the small English Whisper model, which handles clear speech well. Background noise, strong accents and crosstalk reduce accuracy, as they do for any recogniser.
What does “Polish with AI” do?
It loads a small grammar-correction model, also in your browser, and tidies punctuation and sentence structure. It is optional and only downloads when you press the button.
Which voices are available for reading aloud?
Text to speech uses the voices already installed on your device, so the list depends on your operating system and browser.
Does it work offline?
After the models have been downloaded once, transcription and reading aloud both work without a connection.