Why Your Browser's Dictation Tool Needs Wi-Fi But Reading Text Aloud Doesn't

Most browsers can listen to you talk and type it out, and can also read any block of text out loud in a synthetic voice. People tend to assume these are the same trick running backward. They're not. They're two separate halves of the Web Speech API, built on completely different mechanics, and that difference shows up the moment you lose your internet connection.

Two APIs Wearing One Name

"Web Speech API" is actually a spec with two unrelated pieces bolted together: SpeechRecognition, which turns your voice into text, and SpeechSynthesis, which turns text into voice. Browsers shipped them together because they're thematically related, not because they share an engine. Recognition and synthesis solve opposite problems, and each browser vendor implements them with whatever backend makes sense for that direction.

Why Dictation Usually Needs a Live Connection

Turning spoken audio into accurate text is a heavy computation problem -- it benefits enormously from large models trained on huge amounts of speech data, the kind that's impractical to ship inside a browser install. So in Chrome and most Chromium-based browsers, when you use the built-in dictation feature, your microphone audio is streamed to a cloud speech-recognition service for processing, and the transcribed text streams back. That round trip is why dictation can lag slightly behind your speech, why it stops working the moment your connection drops, and why it only runs on secure (HTTPS) pages in the first place -- browsers won't hand microphone access to a page that isn't loaded securely, since raw audio is sensitive data.

It's also why dictation quality varies with background noise and connection speed in a way that feels different from typing. You're not just talking to your computer; you're talking to a remote transcription service through it.

Why Reading Text Aloud Keeps Working Offline

Text-to-speech doesn't have the same bottleneck. Converting written text into audio is a lighter, more deterministic task, and most browsers handle it using voices that are already installed on your operating system or bundled with the browser itself. That's why text-to-speech in your browser usually keeps working with no internet at all -- there's no audio to upload and no transcription model to query remotely. It's rendering locally, voice packs and all.

This Is Also Why Your Voice List Looks Different on Every Device

If you've ever used a browser-based text-to-speech tool on your phone and then your laptop and seen a completely different list of available voices, that's not a bug. The voices available to SpeechSynthesis come from whatever's installed on that specific device -- Windows, macOS, Android, and iOS all ship their own distinct sets of system voices. A website can only offer you the voices your operating system already has. There's no universal voice library a page can reach into; it's reading from local inventory.

What This Means If You're Troubleshooting

If dictation stops working, check your connection and make sure you're on an HTTPS page before assuming your microphone is broken -- those two things cause far more dictation failures than a bad mic. If text-to-speech sounds robotic or limited, that's a function of which voices your OS ships, not something a website can fix on its own; a different device will often sound noticeably better or worse reading the exact same text. And if you need dictation to work somewhere with unreliable internet, it's worth testing ahead of time rather than assuming it'll behave like typing.

Understanding that these are two different systems -- one dependent on a network round trip, one not -- makes the occasional weirdness, the lag, the offline failures, the voice list that never matches across your devices, a lot less mysterious.

You can try both directly in the browser, no account or download required: Speech to Text for dictation, and Text to Speech for having any text read back to you.