Voice Typing
Overview
Choose a local transcription engine or your operating system's built-in dictation service. Availability depends on your platform and which optional engines are installed.
What speech-to-text does
Speech to text turns what you say into editable text wherever Kaiju Hub AI offers a microphone action. Use it to dictate agent prompts, continue a Kaiju Voice conversation, and enter longer instructions without typing. Local engines such as Whisper and Parakeet keep transcription on your machine; platform dictation uses the speech service provided by Windows or macOS.
Available services
Whisper
Local OpenAI Whisper transcription with selectable model size and hardware acceleration.
NVIDIA Parakeet
Fast local speech recognition using the optional Parakeet engine.
Insanely Fast Whisper
An optimized Whisper path for supported GPU configurations.
Windows Voice Typing
Uses Windows dictation and the configured system hotkey.
Apple Dictation
Uses the built-in macOS dictation service.
Whisper
Whisper Settings
Configure OpenAI Whisper settings for local speech-to-text transcription. Whisper runs entirely on your machine for offline, private voice input. Download, install, and remove models directly from this tab.
Available Models
Tiny
39M parameters, ~75 MB download, ~10x realtime speed, requires 1 GB VRAM.
Base
74M parameters, ~145 MB download, ~7x realtime speed, requires 1 GB VRAM.
Small
244M parameters, ~465 MB download, ~4x realtime speed, requires 2 GB VRAM.
Medium
769M parameters, ~1.5 GB download, ~2x realtime speed, requires 5 GB VRAM.
Large
1550M parameters, ~3 GB download, 1x realtime speed, requires 10 GB VRAM.
Turbo
809M parameters, ~1.6 GB download, ~8x realtime speed, requires 6 GB VRAM.
Which Whisper model should I choose?
Start with Small for a practical balance of speed, memory use, and transcription quality. Move down when hardware is limited, or choose Medium, Large, or Turbo when accuracy and GPU performance matter more.
Tiny
Wake-word detection, quick commands, and testing on the lowest-memory systems.
Base
Short notes, prompts, and general dictation on modest hardware.
Small
A balanced everyday choice for agent prompts and longer dictation.
Medium
Accuracy-focused transcription when you can trade extra processing time and memory.
Large
The highest-accuracy local option for systems with at least 10 GB of available VRAM.
Turbo
Fast, high-quality transcription on a supported GPU with at least 6 GB of available VRAM.
Installation Status
The settings page shows the installation status of required dependencies:
Parakeet and Insanely Fast Whisper
Optional fast engines
These engines are separate downloads so the default installation stays small. The settings screen shows installation and hardware status for each engine.
See words stream into your prompt as you speak with live local transcription. Insanely Fast Whisper is optimized for modern NVIDIA GPUs, with Whisper, Parakeet, cloud, and platform speech options also available.
NVIDIA Parakeet
A high-speed local speech-to-text engine. Install its optional runtime from the Voice Typing settings before selecting it.
Insanely Fast Whisper
An accelerated Whisper implementation intended for compatible GPU systems. Install the optional engine before use.
Selecting one through Kaiju Voice never starts a large download silently. Install it in Settings first; the assistant can then verify availability and switch services.
Platform Voice
Windows Voice Typing
Configure Windows Voice Typing for speech-to-text transcription. This service uses Azure Speech Services built into Windows 10 and later. An internet connection is required. Follow the setup instructions to enable speech recognition in Windows Settings.
System Requirements
Windows 10 or later with speech recognition enabled in system settings.
Hotkey
Default trigger hotkey: Ctrl+Shift+W. Configurable to your preference.
Test Trigger
Test button to verify voice typing is working correctly with your microphone.
Apple Dictation
Configure Apple Dictation for speech-to-text transcription on macOS. This is a native macOS feature with no additional setup required beyond enabling dictation in System Settings.
System Requirements
macOS with Dictation enabled in System Settings > Keyboard.
Hotkey
Default trigger hotkey: Ctrl+Shift+D. Configurable to your preference.
Test Trigger
Test button to verify dictation is working correctly with your microphone.
Text to Speech and AI Voices
Local voice models
AI Brain speaks through the selected installed model. Changing models unloads the previous model before loading the new one, and the voice list is filtered to voices compatible with that model.
Kokoro 82M
Lightweight, fast local speech with a broad built-in voice selection.
Chatterbox Turbo
Low-latency expressive speech and supported voice cloning.
Chatterbox
The full Chatterbox model for expressive local synthesis and cloning.
OmniVoice 0.6B
Compact multilingual local text-to-speech.
OuteTTS 1.0 1B
Local speech generation with its own compatible voice set.
Qwen3-TTS 1.7B
Larger multilingual and expressive local synthesis.
Fish Speech S2 Pro
High-quality local speech with supported reference-voice workflows.
Speech speed can be set from 0.5× to 2×. Voice cloning appears only for models that support it; use audio you own or have permission to use.
Cloud provider catalogue
The settings catalogue also includes configuration for OpenAI, ElevenLabs, Cartesia, Smallest AI, Fish Audio, Speechmatics, Amazon Polly, Google Cloud Text-to-Speech, and Microsoft Azure. Availability depends on the provider configuration and credentials you supply.
API credentials remain in Settings and are excluded from Kaiju Voice's spoken or structured settings summaries.