OfflineTTS: Text-to-speech and voice recognition in your browser, no cloud required

A Show HN that delivers: TTS and STT locally, in a tab, for free. We test it.

Tired of letting Google, Microsoft, or OpenAI handle your voice? There's OfflineTTS, a tool that just appeared on Hacker News (July 19, 2026, Show HN by its author).

What it does

Two things, actually:

• TTS (Text-to-Speech) - you type text, it reads it aloud (multiple languages, multiple voices).
• STT (Speech-to-Text) - you speak into your microphone, it transcribes.

The whole thing runs entirely in your browser. Nothing is sent to a server. The page loads the model once (takes a few dozen MB), and then you can use the tool offline - hence the name.

How it works

Under the hood, it's probably transformers.js (HuggingFace's port for browsers), with models compressed into ONNX (Open Neural Network Exchange, a standard format) and quantized. This would explain the reasonable size and the fact that it runs on an average laptop, even without a dedicated GPU.

For whom

• A dev prototyping a voice feature without wanting to pay for a cloud API.
• Someone who wants to dictate a private note without sending their voice elsewhere (lawyer, doctor, journalist).
• A translator who wants instant TTS to check a pronunciation.

Quick verdict

We like: local-first, free, no account to create. Good UX. Works offline after first load.

We temper: the quality of the voices does not yet rival ElevenLabs or premium cloud APIs. For daily use, it's more than enough. For professional audio production, no.

License / price: the tool is free to use. The code is not explicitly labeled on the landing page - check if you want to self-host.

Try it here:offlinetts.com