The last two Junco updates shipped close together, and they change the two things you notice first about the app. One changed how it sounds. One changed how it looks. Here is what happened in each.
Every voice now lives on your device
Up until recently, the audio for your episodes was rendered on our servers, then downloaded to your phone. Version 1.5.0 closed that chapter. Junco now runs Kokoro, an open-weight text-to-speech model, entirely on your iPhone. When a newsletter comes in and gets summarized into a script, your own device turns that script into narration. Nothing about what you listen to passes through another company's cloud on the way to your ears.
The practical differences are easy to feel:
- Episodes are ready the moment they exist. There is no rendering queue on our end anymore, so an episode can start playing as soon as its script is written.
- Offline listening actually works. Once scripts sync to your phone, Junco can narrate them on a plane, in a subway tunnel, wherever.
- Audio is made fast, because it is made nearby. The model downloads once in the background, and after that your phone generates narration at several times realtime.
Since that release we have kept polishing. Version 1.5.1 added resumable renders, so an interrupted episode picks up where it left off instead of starting over, along with faster synthesis and a fix for a memory spike when switching voices.
The bird got a new look
If you updated recently, you also noticed the artwork. We retired the old pixel-art illustrations and redrew the junco as one smooth character with eleven poses. The same bird now loads your episodes, celebrates a streak, wears headphones while you listen, and sleeps when you are done. It shows up in the player, the empty states, and pretty much everywhere else.
We also rebuilt the app icon from the same artwork and refreshed the App Store screenshots. Same junco, sharper feathers.
A quick word on voices
Moving engines meant everyone's default voice reset to Heart. If you have not picked a favorite yet, Settings has the full lineup of eight voices with previews, and it takes about ten seconds. For the backstory on the move itself, see our earlier post, and for the privacy angle specifically, on-device TTS vs the cloud.
FAQ
What is Kokoro?
Kokoro is an open-weight text-to-speech model with 82 million parameters. It is small enough to run comfortably on a phone, and good enough that we deleted our entire server-side audio pipeline. Junco runs it through MLX, Apple's framework for on-device machine learning.
Does Junco's text-to-speech work without internet?
Yes. Scripts sync to your device ahead of time and the audio is generated locally, so playback does not depend on a connection.
Do I need to update to get all of this?
Version 1.5.1 has everything described here, including the new artwork and icon. Grab it from the App Store and you are current.
As always, if something sounds off or looks off, let us know. And if you have not tried Junco yet, download it from the App Store and get your first episode tomorrow morning.