Speaklone Speaklone Download
Back to Blog

Speaklone 1.5 Is in Testing: Free AI Script Analysis, New Avatars, and a Big Stability Fix

· By Dylan Malone
Speaklone 1.5 Is in Testing: Free AI Script Analysis, New Avatars, and a Big Stability Fix

It's been quiet on this blog since February. Not because nothing was happening, but because most of what was happening was the unglamorous kind of work that's hard to write a post about. Speaklone 1.5 is the result, and as of this weekend it's in testing. If all goes well, it reaches the App Store as soon as Apple finishes reviewing it.

Here's what's in it, roughly in order of how much I care about it.

The crash I most wanted gone

Since launch, the most common crash in my reports has shown up under a lot of different names, but underneath it was one bug: a GPU fault deep inside the machine-learning framework Speaklone uses to run its voice models on Apple Silicon. It wasn't something I could fix from my side of the code. The fix finally landed upstream, and 1.5 ships with it.

The upgrade wasn't free, though. The new version of the framework made generation on the Mac about 2.7 times slower, and I wasn't willing to trade a crash for an app that felt sluggish. The cause turned out to be how the work was being split across GPU streams, and I fixed it inside Speaklone, so 1.5 keeps the crash fix and generates as fast as before.

While I was in there, I found a Mac problem I hadn't realized was so bad. When Speaklone wasn't the app in front of you (say you were driving it from the Local API, a hotkey, or a Shortcut with the window hidden, or your display had gone to sleep), macOS's App Nap would throttle it, and generation could run three times slower. Now Speaklone tells macOS it's doing real work while it generates, and goes back to napping when it's done.

AI script analysis, no API key needed

Speaklone's script editor can split a story into narrator and character parts, so each one gets its own voice. The basic splitter runs entirely on your device and handles clean dialogue fine, but for messier prose the AI version is a lot better. Until now, that meant buying Pro or bringing your own Gemini API key.

In 1.5, Speaklone can use Apple Intelligence instead. On iOS 27 and macOS 27 that means Apple's Private Cloud Compute, which Apple designed so your text is used only for that request and isn't stored. Where that isn't available, Pro users get Speaklone's hosted analysis, and everyone else gets Apple's on-device model. Either way, there's nothing to set up.

Since I've written a whole post about on-device privacy, I want to be exact about what leaves your device here. AI analysis only runs when you tap Analyze with AI, and what gets sent is the script text. Your voice, your recordings and your clones never leave your device. If you'd rather keep scripts local too, the regular Analyze Text button still does everything on device.

Avatars, made with Image Playground

Every voice in Speaklone can have an avatar. Apple changed how apps create images on iOS 27 and macOS 27, so 1.5 moves avatar creation into the Image Playground sheet, with a choice of style: Illustrated, Animated, Sketch, or Surprise Me. I also rewrote the descriptions Speaklone builds from a voice for Apple's newer image model, and fixed a few iPhone bugs, including a Regenerate Avatar button that did nothing at all.

Smaller things

  • Shortcuts. A new Speak Text action takes some text and an optional voice and hands back an audio file, so Speaklone can be one step in an automation. It works with Siri too ("Speak with Speaklone"), but honestly, Shortcuts is where it earns its keep. There's deliberately no action for cloning a voice. I want cloning to stay a deliberate, hands-on step.
  • Local API on the Mac. If a client disconnects partway through a request, Speaklone now stops generating right away instead of finishing audio nobody will hear. Failed requests also return the actual error message instead of a generic one.
  • Groundwork for iOS 27 and macOS 27. 1.5 still runs on iOS 26 and macOS 26. A few of the features above need 27.

Testing, and what's next

1.5 is going through TestFlight and App Review now. If you run into a problem once it lands, or you have an idea for what should come next, the Speaklone Discord is the fastest way to reach me, or you can email dylan@dylanmalone.com.

Thanks to everyone who shared crash reports and feedback over the past few months. A lot of this release exists because someone took the time to tell me something was broken.

— Dylan