You need a narration for a short video, or you just want to listen to a long article while doing something else. An online text-to-speech converter handles both cases, but there's a decision upfront that changes the outcome: use your own operating system's voice, or an AI model that runs in the browser?
Two engines, two different purposes
System voices (Web Speech API): the browser asks the operating system to read the text with whatever voices it already has installed. It's instant — no download, no wait — because it doesn't generate an audio file, it plays directly. The limitation is exactly that: since no real file is generated, there's nothing to download. The browser doesn't allow capturing that audio, only playing it.
Local AI (MMS neural model): downloads a ~150 MB speech synthesis model the first time (cached for next time) and generates a real, downloadable .wav file. The voice sounds more natural and consistent because it's the same neural network trained specifically for that language, independent of whatever voices happen to be installed on your computer.
Why "switching voice" sometimes changes nothing
If your system only has one or two voices installed for your language (very common on Windows and Linux without extra voice packs), picking "another voice" from the system selector barely changes anything, because there's genuinely little to choose from. That's the typical scenario where switching to the AI engine is worth it: those models are trained independently of the operating system, with consistent quality no matter what the visitor has installed.
When to use each engine
Use system voices if:
- You just want to listen to the text once, no need to save it.
- You need the result instantly, with no download wait.
- You want to adjust rate, pitch and volume in real time while listening.
Use the local AI engine if:
- You need a downloadable audio file, for example to narrate a video.
- You want a consistent voice that doesn't depend on what the visitor's system has installed.
- The one-time model download (then cached in the browser) is worth the wait.
The AI engine currently supports Spanish, English, French, German, Italian and Portuguese, one model per language.
How the text gets chunked
Browsers silently truncate very long utterances and can stall if the tab has been in the background for a while. That's why the text is automatically split into sentence-sized chunks (roughly 200-300 characters) that are chained one after another, with a short pause between them — the result sounds more natural than one giant utterance, and avoids the typical freezes on long texts.
Frequently asked questions
Why can't I download the audio with system voices? Because the browser only allows playing that synthesis, not capturing it as a file. It's a limitation of the Web Speech API, not of this tool.
Is there a character limit? No artificial cap; the real limit is how long your device takes to process it.
Is the AI model uploaded to a server? No, it downloads once and the text is turned into audio locally, nothing is sent anywhere.
What format do I get with the AI engine? A .wav file.
Listen to or generate audio from any text with text to speech, free and with no sign-up, using instant system voices or downloadable local AI.