Your text to speech voice sounds bad because nobody told you to change it

When a web page asks the browser for the list of available voices, it gets them back in whatever order the system supplies. There is no quality ranking in that list. A tool that simply takes the first entry (which most do), usually lands on a compact system voice, because those tend to sort first and are always present.

So the default is not a considered choice by anyone. It is the first item in an unsorted array. That is the whole explanation for why free browser text to speech has a reputation for sounding cheap.

Apple ships three tiers of the same voice. Compact is the small one installed by default. Enhanced is a larger, noticeably smoother version. Premium is larger again and is genuinely pleasant to listen to for an hour.

They are all free, and you can have as many as you have disk space for.

1. Open System Settings.
2. Go to Accessibility → Spoken Content.
3. Click the System Voice dropdown, then Manage Voices.
4. Expand English (or whichever language you read in). Each voice lists its tiers with a download arrow beside them.
5. Download anything marked Premium. Ava, Zoe, Evan and Nathan are the usual picks. Give it a minute to finish. The progress indicator is easy to miss.

Microsoft's newer voices are labelled Natural and are a substantial jump over the older David and Zira that ship with the system. They are free, they run locally once installed, and there are a lot of them.

1. Open Settings.
2. Go to Time & Language → Speech.
3. Under Manage voices, choose Add voices.
4. Pick your language and install. Anything with Natural in the name is the tier you want. Aria, Guy, Jenny, Libby, Ryan and Sonia are common.
5. Restart your browser so it picks up the new voices.

This is the part worth understanding, because it is usually invisible.

Chrome adds its own entries to the voice list, typically named with a Google prefix. Several of those are synthesised on Google's servers rather than on your computer, which means the text being read is transmitted in order to be spoken. They are often good, and for most reading that trade is fine.

It stops being fine the moment you are reading something you would not paste into a public form: a contract, a medical letter, an unpublished manuscript, anything with your employer's name on it. A tool that does not distinguish between local and online voices cannot honestly make a privacy claim about all of them.

Lumotext groups them and labels the online ones rather than hiding the distinction. If you want the guarantee, pick a voice that lives on your machine.

For reading along, with your eyes on the page and the voice setting the pace, pick a local Premium or Natural voice in your own accent and stop thinking about it. The voice is functioning as a metronome. It needs to be steady and immediate far more than it needs to be beautiful, and a local voice starts speaking with no render step at all.

The very largest neural voices only exist on servers, because no laptop renders them in real time, so a local voice does have a ceiling. What it gives you in exchange is speech that starts the instant you press play, works with no connection, costs nothing per minute, and never transmits a word of what you are reading.

One last thing worth trying once you have a good voice installed: set your speed deliberately, while reading something representative rather than something easy. Most people settle between 1.1× and 1.4× for familiar material and slower for dense material, and almost nobody arrives at that by accident.