Model guide

Local transcription models for Mac

Granite, Parakeet, Qwen3-ASR, or Whisper? Choose by language, speed, timestamps, recording difficulty, and available disk space.

There is no single best speech model

A quiet English call, a multilingual interview, and a noisy recording with domain vocabulary are different problems. LokalBot keeps several engines available so you can choose a fast default and retain a broader or more accurate fallback. All of the options below run locally after their files have downloaded.

ModelBest fitCoverage / size
IBM Granite Speech 4.1 2BRecommended accuracy defaultLocal llama.cpp speech model
Parakeet TDT 0.6B v3Very fast multilingual meetings25 languages; ~190× realtime in project benchmarks
Parakeet TDT 0.6B v2English-focused recallEnglish only
Qwen3-ASR 1.7BHarder multilingual audio52 languages/dialects; ~3.2 GB
Qwen3-ASR 0.6BCompact broad coverageGlobal coverage; ~0.7 GB
Whisper large-v3 turboWide-language fallback and timestamps99 languages; ~1.6 GB

A sensible selection strategy

  1. Start with Granite when you want the project's recommended general-accuracy choice.
  2. Choose Parakeet v3 when throughput and its 25 supported languages cover your meetings.
  3. Keep Whisper installed if you need broader language coverage or word timestamps.
  4. Try Qwen3-ASR 1.7B on difficult multilingual recordings where the compact engines miss too much.
  5. Compare on your audio. A two-minute representative clip is more informative than a generic benchmark.

Speech recognition is only one stage

Speaker separation, punctuation, summary quality, and action-item extraction depend on later stages too. LokalBot begins with separate “Me” and “Them” capture tracks, can apply on-device diarization to the remote side, and then sends the transcript to the selected summarization backend. A perfect language model cannot recover words that the speech model never recognized, so improve capture and transcription before tuning recap prompts.

Storage and memory planning

Speech models are not the only downloads. Local summary and cotyping models range from about 0.53 GB to roughly 17.73 GB in the built-in catalog. Smaller options work on any supported Apple Silicon Mac; several quality-focused choices recommend 16 GB, while the largest long-meeting defaults target 32 GB or more. Install only what you use and keep free space for recordings.

How to evaluate output

FAQ

Questions about this guide

Which model does LokalBot recommend?

IBM Granite Speech 4.1 2B is the current recommended accuracy default in the project documentation.

Which option covers the most languages?

Whisper large-v3 turbo has the broadest listed coverage at 99 languages. Language count alone does not guarantee the best result for a specific recording.

Can I keep several models installed?

Yes. That is useful when your meetings vary by language or audio quality, subject to disk space.

Try it locally

Meeting notes that stay on your Mac by default.

Free, open source, and no account. Apple Silicon · macOS 15.0+

Download for macOS