Local Models
Use Parakeet, Nemotron, Whisper, Cohere, SenseVoice, and Apple Speech on your Mac.
Open AI Models → Local to download and manage transcription models. After downloading, choose a model in Modes → Transcription → Model. The model card menu lets you delete downloaded files or reveal them in Finder.
Local transcription runs on your Mac after setup. Apple Silicon is the recommended hardware; VoiceInk warns that local models do not work reliably on Intel Macs.
Parakeet and Nemotron
| Model | Languages | Real-time |
|---|---|---|
| Parakeet V2 | English | Optional |
| Parakeet V3 | English and supported European languages; used by onboarding | Optional |
| Parakeet Ultra | Multilingual Parakeet variant | Optional |
| Parakeet Unified | English | Optional |
| Nemotron Latin | English, French, German, Italian, Portuguese, Spanish | Required |
| Nemotron Multilingual | Supported languages across multiple scripts, including Arabic, Chinese, Hindi, Japanese, and Korean | Required |
The Mode's language picker shows the actual supported choices. Nemotron models require Real-time for live recording; Parakeet Unified can also be used without it.
Whisper
The predefined Whisper catalog includes:
- Tiny and Tiny (English)
- Base and Base (English)
- Large v2
- Large v3
- Large v3 Turbo
- Large v3 Turbo (Quantized)
Whisper processes the recording after you stop, rather than streaming partial text. Larger models use more storage and can take longer to load and process.
Import a Whisper Model
Use AI Models → Local → Import Local Model… to select a whisper.cpp-compatible .bin file. Imported files are treated as Whisper models and marked multilingual by default. The import is not a general-purpose importer for other model formats.
Choose the imported model in a Mode after importing it.
Cohere Transcribe and SenseVoice
Cohere Transcribe is an on-device transcription model supporting Arabic, Chinese, Dutch, English, French, German, Greek, Italian, Japanese, Korean, Polish, Portuguese, Spanish, and Vietnamese.
SenseVoice Small supports English, Japanese, Korean, Cantonese, and Mandarin Chinese.
Both are downloaded from the Local catalog and process audio without live partial transcripts.
Apple Speech
Apple Speech uses the native macOS Speech framework for on-device transcription and requires macOS 26 or later. It does not offer VoiceInk's real-time streaming option.
macOS manages its language assets. VoiceInk shows download status and can request a language download when supported. Apple limits the number of reserved languages; remove an unused language if you need to make room.
Local Enhancement
The Local tab also contains enhancement options, which process text rather than transcribe audio:
- VoiceInk Refine: Built-in dictation cleanup on a supported Mac.
- Ollama: Connect to a model server, using
http://localhost:11434by default. A local server with a local model keeps enhancement on your Mac. - Local CLI: Run an installed command with VoiceInk's prompt. Commands can use local or hosted services.
See Real-time Transcription and Privacy and Data.