A checking aid, not a narrator
Open an EPUB, DOCX or plain-text manuscript, pick a chapter and a voice, and hear it read aloud, sentence by sentence, while you follow the page. Hearing a text is how a clumsy line, a repeated word or a name spelled two ways gets caught before a narrator finds it in the booth. The model runs in your browser; nothing is uploaded.
Ready.
synthetic voice — not for ACX. The file says so too: its ID3 tag carries
SYNTHETIC_SPEECH=true and the name of the model that made it.
What this is for, and what it is not
Your eye forgives what your ear does not. Hearing a chapter finds the sentence that runs out of breath, the same word three times in a paragraph, the dialogue tag that interrupts a line, and the character whose name is Aldric on page 12 and Aldrich on page 40. It is the pass most authors skip because a person reading aloud takes as long as the book.
Listening to a chapter before the session tells you where the long sentences are and which lines are dialogue before you meet them live. The voice will not show you how to perform it — that is the part that is yours — but it shows you what is on the page.
ACX accepts recordings narrated by a person, and this is a text-to-speech model. Every file this page writes is marked as synthetic inside the file and on the button that saves it. For what each platform says about AI narration, see requirements by platform.
The speech model is Kokoro-82M, about 92 MB, and the engine that runs it about 15 MB more. Both come from this site, not a third party, the first time you press Read it to me, and your browser keeps them for next time. A chapter reads faster than real time on a recent laptop; this page measures your own device’s pace and shows the time left rather than guessing.
Next steps
Names and invented words are where narrators most often have to guess. The pronunciation guide builder finds them in the same manuscript and lets you hear a respelling before you send the guide.
Once a chapter is recorded, proof-listening compares what was said with what is written and lists the places worth listening to again.
Straight answers
No. This is a proofing aid: it reads your text back so you can hear it. ACX's requirements prohibit AI or text-to-speech narration unless it is authorised, and this is not one of the authorised routes. Every MP3 this page saves says it is a synthetic voice, in its file name, on its button and inside the file's own ID3 tag.
No. The manuscript is opened in this browser tab, and the speech model is downloaded from this site and run on your own device; the read-back sends nothing anywhere.
It depends on your device, and the page measures it rather than guessing: after the first sentence it shows the time left. On a recent laptop a chapter is read faster than it takes to listen to. Chapters over 2,500 words are read in parts, so the first part is ready in minutes.
EPUB, DOCX and plain text, or a passage pasted into the box. DRM-protected EPUBs cannot be opened. PDFs are not read here, because a PDF keeps page layout rather than chapters; export the manuscript to DOCX or EPUB instead.
Source: the ACX audio submission requirements. What every platform says about AI narration is compared on requirements by platform.