Speech engines
| Mac | iPhone and iPad | |
|---|---|---|
| Speech to text | Whisper (whisper:base, whisper:small) and Moonshine streaming (moonshine:tiny-streaming, small-streaming, medium-streaming). | Apple speech recognition. |
| Text to speech | Kokoro (12 voices), your cloned voice, or macOS system voices. | iOS system voices. |
On the Mac, speech models are ordinary catalogue downloads in Settings › Models, and voices are in Settings › Voice › Neural voices. All of them run on the Mac. The Moonshine models are English only. Kokoro and voice cloning speak English.
Voice mode
Voice mode is a hands-free conversation: you speak, Kalon replies aloud, and it listens again.
- On the Mac, voice mode uses Moonshine when it is installed, for live captions while you speak, and Whisper otherwise.
- By default Kalon listens continuously and detects when you stop. Mic sensitivity sets how softly you can speak.
- With Push to talk on, hold the microphone button or the Space bar to speak and release to send.
- Voice starts a new chat begins a fresh conversation for each session instead of continuing the open one.
- The camera button shows a preview of yourself. Frames are not sent to the model.
- Press Esc or click End voice mode to finish.
These settings are in Settings › Voice. On iPhone, start voice mode with Talk hands-free in the composer. Tap the microphone in the control pill to pause or resume listening.
Online realtime voice
On the Mac, if you have turned on online models and selected an online model from OpenAI, xAI or Google, voice mode uses that provider's own speech-to-speech service instead of the local pipeline. Your microphone audio and Kalon's standard instructions are streamed to the provider for the length of the call, and the reply is spoken in the provider's voice. With online models off, or a local model selected, voice mode stays on the Mac. See Privacy and network.
Dictate anywhere (Mac)
Press ⌃⌥Space in any app to dictate into it. A floating pill appears over the app you are using, and it never takes focus away.
- Press ⌃⌥Space and speak.
- Press Return to insert the text at the cursor, or Esc to discard it.
- Kalon transcribes with Whisper on the Mac and pastes the text into the app that was focused.
Requirements:
- A Whisper model. Without one, the pill says "Install a Whisper model in Kalon → Models to dictate".
- Microphone access.
- Accessibility access, so Kalon can paste for you. Without it, Kalon leaves the text on the clipboard for you to paste with ⌘V and asks macOS to show the permission.
Kalon saves your clipboard before it pastes and restores it afterwards. Dictation is refused in password fields. Each take is limited to 120 seconds.
Turn the shortcut off with Dictate anywhere in Settings › Voice › Voice mode. It is on by default, and the pill only appears when you press the shortcut.
The microphone button in the composer dictates into the message you are writing. It also uses Whisper.
Reading replies aloud
Choose a voice with Spoken voice in Settings › Voice › Speech. Best available picks one for you. The play button beside it plays a sample. Click Read aloud under any reply to hear it.
On iPhone, choose a voice under Settings › Voice. More system voices can be added in the iOS Settings app.
Cloning your voice (Mac)
Kalon can clone your voice on the Mac with the Qwen3 TTS voice pack.
- In Settings › Voice › Neural voices, download Qwen3 TTS.
- Under Your voice, choose Record a new voice.
- Read the passage aloud. It takes about fifteen seconds. A take must be between 10 and 30 seconds.
- Name the voice and click Create voice.
- Select it under Spoken voice.
The recording and the voice profile are stored in Kalon's data folder on the Mac. Delete removes both.
Speech on iPhone and iPad
The phone uses Apple's speech recognizer and asks it to recognize on the device. If on-device recognition is not available for your language, iOS may process the audio with Apple's speech service instead.