Reviewed 4 November 2026 · ROP product team · Draft
Quick answer
The family reader offers up to three installed local English voices, preferring Daniel when available. The separate natural-voice desktop package bundles Kokoro with Heart and Emma. Test the voice on your own device before judging its quality.
Know which reader you opened
The family desktop and mobile readers use installed device voices. Their availability and quality vary. The natural desktop experiment includes its model and Heart and Emma voices. Passing a device-voice test does not prove that Kokoro works in the extension.
Compare a short passage at 1×
Use the same public paragraph for each voice. Include a normal sentence, a name and a number. Begin at 1×, then adjust speed only if the words are comfortable to follow. A familiar sentence alone may hide pronunciation problems.
If no voice appears
Wait for the reader to finish loading voices, then reopen it. Check that your device has an installed English voice. Remote and novelty voices are excluded from the family shortlist. Daniel is preferred when present, but is not required. Do not assume that a voice installed on one device exists on another.
If the natural package is slow
The first load can differ from a warm run. Open Test diagnostics in the natural desktop reader and record load time, estimated first word, device and thread count. Confirm the first word you actually hear with a stopwatch. Real Mac quality and ten-minute playback are still awaiting acceptance.
Share useful feedback
Report the package version, device, browser version, selected voice, first-word seconds and a comfort rating from 1 to 5. Note any gap or crash. Do not paste the passage or private account details into feedback.
Common questions
Is the sample the same as Kokoro?
The website sample uses installed device voices. It is not evidence of Kokoro performance.
Will every device sound identical?
No. Device voices differ, and the bundled natural-voice experiment still needs real-device verification.
Keep exploring
Explore the labelled voice sampleThe website sample uses device voices, not the Kokoro experiment.