Transcription
Canary ASR
Local Canary speech recognition on Apple silicon, with the Sophea model for Greek and English and import of compatible models.
About
Canary ASR runs Canary speech recognition models on your Mac through MLX. The add-on offers one model for download, Sophea Canary for Greek and English, and imports further compatible Canary models.
This is a local engine. Your audio is transcribed on your Mac. The add-on uses the network to download a model.
TypeWhisper for Windows has no Canary ASR add-on. There the model Canary 180M Flash is part of Local Models (sherpa-onnx).
Requirements
This release requires TypeWhisper 1.7.0 or newer on macOS 14 or later and a Mac with Apple silicon.
Sophea Canary is a download of about 1.9 GB. The add-on states 8 GB of memory or more for it. You need no API key and no account.
Setup
- Install and enable Canary ASR in Settings > Integrations > Discover.
- Open the add-on settings and select Download & Load beside Sophea Canary (Greek / English).
- Select Canary Speech (MLX) as your transcription engine in the dictation settings.
- Set the language of the dictation to Greek or English. Canary does not detect the language on its own.
Settings
| Setting | What it controls |
|---|---|
| HuggingFace Token | Optional. A token raises the download rate limits of Hugging Face. Save checks the token with Hugging Face and stores a valid one in the macOS Keychain. Remove deletes the stored token. |
| Model | Lists Sophea Canary and your imported models with their size and memory requirement. Download & Load downloads Sophea Canary and loads it. Unload releases the model from memory; for Sophea Canary it also deletes the downloaded files. |
| Import Model… | Adds a Canary model from a Hugging Face repository or from a local folder with config.json, tokenizer files, and .safetensors weights. The add-on checks and loads the model before the import completes. |
An imported model is a copy that TypeWhisper keeps; your original files stay unchanged. A token that you enter in the import dialog for a private or gated repository is used for this import and is not saved.
Transcription
The add-on transcribes the finished recording. It does not stream partial text while you speak. It splits a longer recording into sections of about 20 seconds at a pause nearby and transcribes them one after the other.
Sophea Canary recognizes Greek and English. Without a selected language the add-on stops with a message that asks you to choose one. Translation is not supported, and TypeWhisper dictionary terms are not passed to the model.
Local processing and privacy
Recognition runs on your Mac. The add-on sends no audio and no text to a cloud service.
The files of Sophea Canary come from the Hugging Face repository KIEFERSA/Sophea-Canary-ASR-mlx: weights, configuration, and tokenizer. TypeWhisper stores them in ~/Library/Application Support/TypeWhisper/PluginData/com.typewhisper.canary/models, imported models in custom-models beside it. The add-on also contacts Hugging Face when it checks a token and when you import a model from a repository.