01
Record or import audio
Start a meeting, interview, or class recording, or bring in existing audio from Files and the share sheet.
How it works
Follow it live, then finish with the complete result.
HushNote brings mixed-language live transcription, real-time speaker labels, transcript refinement, and final speaker calibration into one local-first path.
02
Transcribe mixed speech live
Chinese, English, and natural switching stay in one low-latency stream you can read during recording.
03
Label speakers in real time
Designed for 2–4 people. Speaker labels catch up as the model becomes ready, while ASR remains the highest priority.
04
Refine both layers after recording
A stronger model corrects transcript segments and the full recording calibrates the speaker timeline before summary and export.
Product boundary
What HushNote intentionally does not try to be.
- Not a team workspace with cloud collaboration layers.
- Not a chat-first AI assistant with a giant prompt box.
- Not a heavy setup flow that asks you to configure before you can capture.
Built for
Focused recording scenarios where privacy and readability matter.
Meetings
Capture the live discussion, then scan the structured summary and the key moments afterward.
Interviews
Keep the interview flowing while HushNote helps you return to the important sections later.
Voice notes
Use the same workflow for solo capture when you want a cleaner review than a plain audio list.
Sensitive notes
The interface and runtime both stay aligned with a local-first, privacy-conscious posture.
What changed
Version 1.3.0 gives live and final results distinct jobs.
The live path prioritizes latency and readability. The post-recording path uses full context to improve transcript accuracy and speaker attribution.
Mixed speech stays in one stream
The new live engine reduces breaks around language switching and keeps bilingual speech in one transcript.
Low-latency live labels
Speaker processing runs beside ASR and yields whenever the highest-priority transcription path needs room.
Full-audio final calibration
After recording, HushNote checks the complete speaker timeline. If a split cannot be mapped reliably, the original ASR segment stays intact.
Summary models no longer inflate the initial install
Download a local summary model only when needed, choose one that fits the device, and remove it at any time.
Start with the simplest version
The point is not to add more screens. The point is to help you capture once and review clearly.
If that matches how you work, the App Store version is already live for iPhone.