Whisper Models

Choose the right model for your needs

When using local transcription, you need to download a Whisper model. This guide helps you choose the right one.

Available Models

ModelSizeSpeedAccuracyBest For
tiny~75 MBFastestBasicQuick tests, low-power devices
tiny.en~75 MBFastestBasicEnglish-only, slightly better
base~150 MBFastGoodGeneral multilingual use
base.en~150 MBFastGoodRecommended for English
small~500 MBMediumBetterWhen accuracy matters
small.en~500 MBMediumBetterProfessional English dictation
medium~1.5 GBSlowHighTechnical/medical content
medium.en~1.5 GBSlowHighHigh-stakes English content
large-v3~3 GBSlowestHighestMaximum accuracy needed

English vs Multilingual

Models ending in .en are English-only. They’re:

  • Faster — Optimized for a single language
  • More accurate — No confusion with other languages
  • Recommended — If you only dictate in English

Multilingual models (without .en) support many languages but may be slightly less accurate for any single language.

Downloading Models

  1. Go to Settings in Yap
  2. Scroll to the Models section
  3. Click on a model to download it
  4. Wait for the download to complete

Models are downloaded from HuggingFace and stored in:

~/Library/Application Support/yap/models/

Changing Models

  1. Ensure the desired model is downloaded
  2. In Settings, click on the model to select it
  3. The green checkmark indicates the active model

You can have multiple models downloaded and switch between them as needed.

Model Performance

Apple Silicon (M1/M2/M3)

ModelTranscription Time (10s audio)
tiny.en~1 second
base.en~2 seconds
small.en~4 seconds
medium.en~8 seconds
large-v3~15 seconds

Intel Mac / Windows

Expect 2-3x slower performance compared to Apple Silicon.

Storage Considerations

Models take up disk space:

Models DownloadedTotal Storage
tiny + base~225 MB
All English models~3.2 GB
All models~7 GB

You can delete models you don’t use to save space:

# View downloaded models
ls ~/Library/Application\ Support/yap/models/

# Delete a specific model
rm ~/Library/Application\ Support/yap/models/ggml-base.en.bin

Recommendations

Daily Use

base.en — Best balance of speed and accuracy

Professional Work

small.en or medium.en — When accuracy is critical

Low-Power Devices

tiny.en — Fast enough for real-time feel

Technical Content

medium.en or large-v3 — Better with jargon and specialized terms

Non-English Languages

small or medium (multilingual) — Good accuracy across languages