whisper.cpp (Local)

About

whisper.cpp (Local) adds offline Whisper transcription to TypeWhisper for Windows through Whisper.net. It downloads GGML model files, loads the selected local model, and can transcribe without sending audio to a cloud service.

Features

• Offline Whisper transcription on Windows
• Uses Whisper.net with whisper.cpp runtime files
• Built-in GGML model downloads
• Download, load, unload, and remove controls for local models
• Automatic language detection
• Prompt support for transcription context
• Translation support
• Automatic, CPU, NVIDIA CUDA, AMD Vulkan, and custom AMD ROCm acceleration options
• Runtime diagnostics with restart requirements and explicit CPU fallback reasons
• No API key or cloud account required

Transcription Models

Model family	Variants
Tiny	multilingual, English-only, Q5_0
Base	multilingual, English-only, Q5_0
Small	multilingual, English-only, Q5_0
Medium	multilingual, English-only, Q5_0
Large V3 Turbo	full and Q5_0

Recommended Models

Model	ID	Size
Base	base	~142 MB
Base (Q5_0)	base-q5_0	~57 MB

Configuration

• Model - Choose the Whisper model to download and load.
• Prompt - Optional context can be passed into the transcription processor.
• Translate - Enables Whisper translation when requested by TypeWhisper.
• Acceleration - Use automatic selection, CPU, NVIDIA CUDA, AMD Vulkan, or a custom ROCm whisper.dll path.

Setup

1. Open TypeWhisper Settings > Integrations
2. Select whisper.cpp (Local)
3. Download and load your preferred model
4. Select whisper.cpp as your transcription engine in Settings or in a profile

Notes

• Models are stored under the plugin data directory.
• Quantized Q5_0 variants reduce download size compared with full models.
• Larger Whisper models can improve accuracy but require more disk space and memory.
• CUDA can be installed by TypeWhisper; Vulkan requires a compatible AMD driver, while ROCm uses TYPEWHISPER_WHISPERCPP_ROCM_LIBRARY_PATH.