Spoken vs Speechify, NaturalReader and ElevenReader

If you want Chrome to read articles, docs and PDFs aloud, Spoken is not your only option. Speechify, NaturalReader and ElevenReader are all established tools, and each does some things better than we do. This post compares the four as of November 2026, using each company’s own pages. The sources are listed at the end. Chrome and Edge also have built-in read aloud; we compare those in a separate post.

The short version: Spoken runs the voice model on your own machine, so the text you listen to stays in your browser. Speechify and ElevenReader process your text in the cloud, and NaturalReader does for its paid AI voices (its free tier uses your device’s system voices). In exchange they offer more voices, more languages and apps on your phone.

At a glance

	Spoken	Speechify	NaturalReader	ElevenReader
Where text is processed	In your browser, on your GPU (WebGPU) or CPU (WASM fallback)	In the cloud; on-device voices exist on iOS only	Free voices are your device’s system voices; Plus voices are Microsoft Azure voices and Pro voices are powered by Gemini and ChatGPT models (pricing page), so paid AI voices run in the cloud	In the cloud; the extension imports the page into ElevenReader
Account	None. Spoken sends no analytics or telemetry. It downloads the model from Hugging Face once, and for PDFs it re-fetches the file from the site you are already viewing	Sign-in offered in the extension; we couldn’t confirm if free use needs one	Not clearly stated for the extension	Required
Price	Not yet available to install; no price announced. Synthesis uses your own hardware, so there are no API costs behind it	Free (10 voices, up to 1.5x); Premium $29/month or $139/year	Free; Plus $119/year or $20.90/month; Pro $159/year or $25.90/month	Free (10 hours/month); Ultra $8.25/month billed annually
Offline after download	Yes, once the model is cached	Premium (offline listening is listed as a Premium feature); on-device iOS voices also work offline	Paid plans, by exporting MP3s	Ultra: 150 hours of offline downloads a month, each kept up to 60 days and renewable
Voices / languages	21 voices, English only (US and UK)	400+ voices on Premium (1,000+ per its extension page), 60+ languages	150+ voices, 20+ languages (Chrome Web Store listing); 90+ languages with Plus voices (pricing page)	90+ languages, included free (its pricing page; a press item on the same page says 32)
PDF	Yes, in Chrome’s and Edge’s built-in viewer, up to 100 MiB	Yes, plus EPUB, DOCX, XLSX and scanned pages	Yes, plus Kindle, Google Docs and email	Yes, plus EPUB, DOCX, TXT and HTML (up to 500 pages); uploads count toward the 10 free hours

How Spoken works today

Spoken runs Kokoro-82M, an open speech model, inside the extension. On a machine with WebGPU it uses your graphics card. Where WebGPU isn’t available, it switches to a WebAssembly build on your processor, and the engine status says which one is running.

The first time you use it, Spoken downloads the model from Hugging Face: about 326 MB for the WebGPU version, or about 92 MB for the WASM fallback. The status line shows the real download progress in MB. After that the model is cached, and there is no analytics or telemetry. The engine status reads “Nothing leaves this browser.”

You can start reading in a few ways: select text and use the on-page pill, right-click and choose Speak selection, press Alt+Shift+S (Alt+Shift+K on ChromeOS), or press Start on the “Listen to this article” pill. The current sentence is highlighted on the page as it reads, along with an estimate of the current word. It works in Chrome 125 and later.

Where each competitor is the better pick

Speechify

Pick Speechify if you want one library across every device. It runs on iPhone, iPad, Android, Mac, Windows, the web, Chrome and Edge. Its Premium plan has a far larger voice catalog than ours and more than 60 languages, playback up to 5x (Spoken tops out at 2x), and Scan & Listen for printed pages. It also reads EPUB, DOCX and XLSX files and adds AI summaries, chat and podcasts. Since December 2025 Speechify has offered on-device voices on iOS that run fully on the phone; its announcement doesn’t say which plans include them.

NaturalReader

NaturalReader is a mature extension with a generous free tier: unlimited listening with free voices, which are your device’s own system voices. Its Plus voices cover more than 90 languages, according to its pricing page, it can export MP3s for listening anywhere, and it reads Kindle books, Google Docs, Gmail and Outlook. It has web and mobile apps and an OCR scanner on Plus. NaturalReader also states plainly that it doesn’t show ads, doesn’t sell user data to advertisers and doesn’t use your content to train AI models.

ElevenReader

ElevenReader gives you ElevenLabs voices, which many listeners find more expressive than a small on-device model, and a library that follows you from Chrome to your phone. Synthesis runs on their servers, so it works on low-end devices with no GPU and no model download. The free plan includes 10 hours of audio a month, and the free Explore library of audiobooks doesn’t count against it.

In general, cloud voices can sound more expressive than a small on-device model, and every one of these products speaks more languages than Spoken does today.

Where Spoken has an edge: privacy and cost

Privacy

With Spoken, the text you listen to is turned into audio on your machine. It isn’t uploaded to read it aloud.

The others work differently. Speechify’s privacy policy says submitted information is processed and stored in the United States, and that Microsoft Azure or Amazon Polly receive the text when you pick one of their voices; its own voices run on cloud infrastructure (Baseten), except the on-device iOS voices. ElevenReader’s extension imports the page’s full HTML into your ElevenReader library (its help center notes this works for paywalled pages you’re logged in to), so the text is processed by ElevenLabs. The ElevenLabs privacy policy, which covers its websites and mobile apps, says it uses personal information to train its AI models; you can opt out under Data use in account settings, for future training. NaturalReader’s free system voices can work without its cloud, while its Plus and Pro voices are Azure, Gemini and ChatGPT-powered. None of this is hidden; it’s how cloud speech works. If you read medical notes, legal documents or internal work pages, that difference may matter to you.

Cost

Because synthesis runs on your own hardware, there are no API costs behind Spoken. You don’t have a character quota, a monthly hour allowance or downloads that expire. NaturalReader’s free system voices are also unlimited; the quotas apply to its AI voices and to ElevenReader’s and Speechify’s cloud voices. The trade-off with Spoken is the one-time model download and the use of your own GPU or CPU.

How Spoken handles latency

Spoken doesn’t wait on a network round trip for each sentence. It splits text into sentences and synthesizes the next one while the current one plays, with a shorter first chunk so audio starts sooner. We haven’t published latency measurements, and how fast it feels depends on your hardware. On a slow machine without WebGPU, a cloud service may start faster.

What Spoken doesn’t do yet

To be clear about the limits: Spoken is English only, with 2 British voices among the 21. It has no mobile app, no MP3 export, no OCR and no seek bar. It doesn’t yet have its own cleanup for URLs, numbers and acronyms, so a URL is spelled out character by character and some acronyms are read letter by letter. It runs best on a machine with WebGPU, which uses a one-time download of about 326 MB. Without WebGPU it falls back to a slower CPU engine (about 92 MB). And it isn’t available to install yet.

If those limits don’t get in your way, and you’d rather your reading stay on your own computer, Spoken is worth a look when it launches for Chrome.