Local models
HushMemo separates speech-to-text models from summary models. Transcription models turn audio into text; summary models turn reviewed transcripts into notes, TODOs, emails, and templates.
Transcription
Choose a transcription model based on storage, speed, and accuracy. Smaller models are faster; larger models can handle more complex recordings with better results.
Fastest transcription
Small speech-to-text model for quick drafts and short voice notes.
Best for: Quick thoughts, short reminders, rough drafts, and devices with limited storage.
Fast and compact, but less accurate with noise, accents, long meetings, or domain vocabulary.
Balanced transcription
Balanced local transcription model for everyday recordings.
Best for: Meetings, calls, lectures, interviews, and voice notes where speed still matters.
A practical default after download. Important transcripts should still be reviewed before summarizing or sharing.
Higher accuracy
Higher-accuracy local transcription model for longer and messier audio.
Best for: Longer meetings, lectures, interviews, mixed speakers, and recordings with more detail.
Uses more storage and processing time. Long recordings may still need section-by-section review.
Summaries
After transcription, local summary models organize the reviewed text into structured outputs such as meeting notes, tasks, emails, and study notes.
Default recommended
Balanced local summary model for everyday notes.
Best for: Meetings, quick summaries, TODO extraction, and short-to-medium transcripts.
Runs locally after download. Long recordings may need shorter summaries or chunked review.
Chinese preferred
Higher-quality Chinese summaries with a larger context window.
Best for: Chinese meetings, consultation notes, and templates with more structure.
Optional manual download. Larger file size means more storage and longer processing time.
High quality
Opt-in high-quality model for longer context and complex templates.
Best for: Longer transcripts, deeper speech analysis, and multi-section templates.
Optional manual download. It can be slower and should be used on devices with enough free storage and RAM.
Offline boundary
Network access is mainly for model downloads, store licensing, and explicit website link-outs. After a model is downloaded, transcription and summaries run locally unless you export or share content yourself.
