I Tested 7 TTS Chrome Extensions. Most of Them Disappointed Me.

You know what started this whole thing? OpenClaw. Not because OpenClaw is a text-to-speech tool — it isn't, really — but because when it exploded in February, quarter million GitHub stars, lines outside Tencent HQ in Shenzhen, the whole circus, everybody on Reddit started asking the same question. Can this thing read articles to me?

And yeah, it can. You spin up Docker, wrestle with YAML configs, grab an API key from ElevenLabs or OpenAI, and forty-five minutes later your computer speaks. For reading a blog post. Forty-five minutes. There are Chrome extensions that do this in three seconds flat, and nobody seems to know they exist. (If you're not sure what TTS actually means, here's a plain-English explainer.)

So I installed seven of them. All at once, running side by side for two weeks. Same three test pages through each one — a 4,000-word Substack essay about housing policy, a Wikipedia article on the Ottoman Empire with tables and footnotes and those little bracketed citation numbers everywhere, and a New York Times piece with a soft paywall and aggressive ad placement. What follows is brutally honest.

Okay — CastReader first. Disclosure. This is our product. Take everything here with a grain of salt the size of a grapefruit. Why did we build it? Because I kept hitting play on other TTS extensions and hearing TRENDING NOW, SIGN UP FOR OUR NEWSLETTER, RECOMMENDED FOR YOU before the actual article started. Every single time. The extraction problem drove me up the wall. That's the itch CastReader scratches — it reads the DOM, figures out where the article body lives, throws away the garbage, and starts reading the actual content. On that NYT page? Nailed it. Navigation bar, gone. Sidebar ads, gone. More from the Times footer block, gone. Just the article.

The paragraph highlighting is the other thing I'd fight for. Current paragraph lights up, page auto-scrolls. I started using it while making dinner — glance up from the cutting board, see exactly where I am, look back down. Stupid simple. Weirdly useful. What's not great — the voice library is modest compared to Speechify, there's no iOS app, no Android app, it's Chrome and that's it, and our user base could fit in a mid-size conference room. Free tier is legitimate though. Not a free trial that expires Tuesday. Real free. You can try it right now. It works especially well on platforms like Kindle Cloud Reader and Medium, where other extensions tend to choke on the page structure.

Now Speechify. The voices are incredible. I need to say that first because it's true and pretending otherwise would be dishonest. Their premium neural voices sound better than anything else on this list. Natural pauses. Handles em-dashes gracefully. Doesn't butcher proper nouns as often as you'd expect.

That'll be $139 per year, please.

The free version gives you I think three voices? Maybe four? They sound like they were recorded during the Obama administration. And the usage limit is tight — I blew through it on a single long read. Here's my actual problem with Speechify though, and it has nothing to do with the product itself. Go to their blog. Right now. Open a new tab. You'll find Anne Rice Books in Order. Wolf Meme. How to Watch the Hannibal Movies in Order. Garage Sale Mysteries in Order. Hundreds of articles. None of them have anything to do with text-to-speech. It's an SEO farm. And it's a successful one, these pages rank, but it tells me something about where the company's attention goes. Are they improving the extension, or are they writing their 47th Books in Order article? The desktop app uses 800MB of RAM sitting idle. Make of that what you will.

Read Aloud is the one the nerds love and I mean that as a compliment. Open source. Free. A million users. Connects to basically every voice engine on planet Earth — Google WaveNet, Amazon Polly, IBM Watson, Microsoft Azure, OpenAI. Plug in your API key and you're running premium voices at API cost which is pennies per article. Without an API key though you get Web Speech API. On macOS, tolerable. On my work ThinkPad running Windows 11... imagine a Speak & Spell from 1978 trying to pronounce geopolitical. That bad. I counted the settings panel. Twenty-three options. My colleague, ships code for a living, smart person, looked at it and said nope and closed it. No paragraph highlighting either. Word-level exists but it jitters — lands on the wrong word, corrects itself, jitters again. For developers with cloud accounts it's genuinely the best option. For literally anyone else it's way too much friction.

NaturalReader. Been around forever. Not flashy. Gets the job done like a reliable uncle at Thanksgiving. Ships a couple decent AI voices in the free tier, not Speechify-gorgeous but leagues better than browser defaults. I used the Aria voice for three straight days reading research papers and it was... fine? Comfortable? Like a well-worn pair of shoes that you don't think about. The immersive reader mode strips away page design and gives you black text on cream background — used this on a few cluttered news sites and the experience improved a lot. Oh and the dyslexic font toggle, switches to OpenDyslexic, I sent the extension to my cousin who has dyslexia and she texted back "wait where has this been." Not a revolutionary feature. A deeply considerate one. Pricing is one-time at $99.50 which I respect in a world of subscription fatigue but you're buying the whole platform not just the Chrome extension. The UI needs work — switching voices requires clicking the icon, opening a panel, scrolling to voices, selecting one, closing the panel. Four steps for something that should be a dropdown.

Talkie. Zero cloud. Zero servers. Zero data leaving your laptop. Runs on Web Speech API, your operating system's built-in voices, processed locally, nothing goes anywhere. Code is on GitHub. Payment model is pay what you want which includes zero. I tested Talkie while reading internal company documents, stuff I absolutely would not paste into a cloud-based TTS. macOS Siri voices through Talkie — acceptable, maybe a six out of ten. Windows default voices — painful, four out of ten on a generous day. No highlighting of any kind, you press play and hope you can keep up. If you work with sensitive or confidential text and need TTS, Talkie is the only responsible choice. Otherwise move along.

I almost didn't include Snap&Read because it's really a K-12 education tool. But the text leveling feature is wild enough to mention — you're reading a graduate-level article, toggle the slider, suddenly it's rewritten at an eighth-grade reading level while being read aloud. The underlying information preserved but vocabulary and sentence complexity simplified in real-time. I have zero use for this personally but my friend teaches ESL to adults and when I showed her she literally said are you kidding me and pulled out her credit card. $4/month, most schools buy bulk licenses so students get it free.

Capti Voice. Built with NSF and NIH funding. You can tell because it has that unmistakable designed-by-researchers aesthetic — functional, dense, lots of small buttons, menus nested inside menus. The concept is a reading playlist, queue up articles and documents and web pages, listen in order. Like a podcast you assembled yourself from your reading backlog. Sounds brilliant but in practice when I just want to hear one article right now, adding it to a playlist first feels like being asked to create a Jira ticket before I can start actual work. Word highlighting is solid, twenty-six languages, but you have to register an account which is one more barrier than I want between me and hearing an article read aloud.

So about OpenClaw — the thing that started all this. It's extraordinary for what it actually does, orchestrating multi-step tasks, sending emails, writing code, managing files. General-purpose AI agent and a genuinely good one. For reading web pages aloud though it's the wrong tool completely. Chrome extensions live inside your browser tab, they see the page structure directly, they highlight text on the actual page, they start speaking in milliseconds because there's no round-trip to a separate process. OpenClaw lives in its own container, fetches page content through an HTTP request separate from your browser, processes it through an LLM, pipes text through a TTS API. The result is audio with no visual connection to the page you're looking at. And it took me 45 minutes to set up what a Chrome extension does in three seconds. Different tools, different jobs.

Who wins? Depends what bothers you most. Bad page extraction, hearing menu items and ad copy? CastReader handles that best, yes I'm biased but it's also true. Voice quality above everything and money's no object? Speechify and it's not close. Maximum technical flexibility with your own API keys? Read Aloud. Reading confidential stuff? Talkie, only real option. Teaching or learning? Snap&Read or Capti Voice. Start free — CastReader, Read Aloud, and Talkie cost nothing and don't expire. Try one for a week. You'll figure out pretty fast whether you need something else.