User guide
Install, set up, choose Hush AI Microphone in your apps, and fix problems.
Install#
Download the file for your computer from the latest release.
| Computer | File | First start |
|---|---|---|
| Windows 10 or 11 | …_x64-setup.exe (ARM: …_arm64-setup.exe, NVIDIA RE-USE only for now: DeepFilterNet 3's inference library does not build for Windows on ARM yet) | SmartScreen: More info › Run anyway |
| Mac with Apple silicon | …_aarch64.dmg | Right-click the app › Open (or System Settings › Privacy & Security › Open Anyway) |
| Intel Mac | …_x64.dmg | Same as above |
| Linux | .AppImage, .deb or .rpm | Needs PipeWire or PulseAudio (pactl) |
The warnings appear because the apps are not code-signed yet. They appear only once.
First start#
- Voice model. Choose a model and download it. Hush checks the file before it uses it.
- Efficient (DeepFilterNet 3, 8 MB, recommended): little CPU, MIT or Apache-2.0 license, so you can use it at work too.
- NVIDIA RE-USE (54 MB): more CPU. Agree to its license terms first.
- Microphone. Choose your microphone and say something. The bar moves when Hush hears you.
- Hush AI Microphone.
- macOS: press Install. macOS asks for your password once.
- Windows: Hush shows four steps: download the VBCABLE_Driver_Pack from vb-audio.com/Cable, extract it, run VBCABLE_Setup_x64 as administrator, press Install Driver, restart Windows. Open Hush again and press Rename. Windows asks for permission once.
- Linux: nothing to do.
- Hear the difference. Record six seconds with some noise around you (type, tap the desk), then play Original and With Hush.
- Use it. Choose Hush AI Microphone in your apps.
Choose Hush AI Microphone in your apps#
| App | Where |
|---|---|
| Zoom | Settings › Audio › Microphone |
| Google Meet | More options › Settings › Audio › Microphone |
| Microsoft Teams | Settings › Devices › Microphone |
| Discord | User Settings › Voice & Video › Input Device |
| Slack | Preferences › Audio & video › Microphone |
| OBS Studio | Settings › Audio › Mic/Auxiliary Audio |
| Browser apps | The browser's site settings, or the app's own audio settings |
| Windows | Settings › System › Sound › Input |
| macOS | System Settings › Sound › Input |
Do not choose Hush AI Microphone as your microphone inside Hush. Hush hides it from its own list.
Every day#
- The notch. Point at the island at the top of the screen: noise cancellation, Live Translate, their voice, captions and your next meeting.
- Tray or menu bar. Click the mochi to open the notch (or the widget, if you chose it). Turn noise cancellation or Live Translate on or off, swap languages, change the microphone, or join your next meeting. Double-click opens the window (Windows). Right-click for the menu.
- Closing the window keeps Hush running in the tray. Use Quit to stop it. When Hush is not running, Hush AI Microphone is silent.
- Open at login (Settings) starts Hush in the tray when you log in.
Voice model#
| Model | Use it for | Cost |
|---|---|---|
| Efficient (DeepFilterNet 3) | Everything (48 kHz full band) | Lowest CPU, one thread |
| NVIDIA RE-USE | Light, Balanced or Studio, below | More CPU |
Switch on the main screen. Settings › Voice models downloads or removes each model. On the VoiceBank-DEMAND test, DeepFilterNet 3 scored higher than RE-USE Light on every metric.
RE-USE quality#
| Quality | Use it for | Cost |
|---|---|---|
| Light | Calls and meetings (16 kHz voice band) | Lowest delay and CPU of RE-USE |
| Balanced | Clearer voice (24 kHz) | More CPU |
| Studio | Recording, streaming (48 kHz full band) | Needs a fast CPU |
The Load chip shows how much of real time the model needs. Below 100% it keeps up. If it stays high, choose Efficient, or Light with RE-USE.
Studio: clean up a recording#
Open the Studio tab.
- Drop a file on the window, press Choose a file, or press Record yourself. Pick the microphone to record (the bar shows its level) and press Start recording (up to 30 minutes). Hush records the microphone as it is and cleans the recording afterwards; your calls and Hush AI Microphone are not affected.
- Hush cleans it with your voice model. Long files are fine; the progress ring shows the time left.
- Press play and switch between Original and With Hush while it plays (the A key switches, Space plays).
- Save clean WAV writes a 48 kHz mono file. Show in folder opens where you saved it.
Files: WAV, MP3, M4A/AAC, FLAC, OGG, AIFF, CAF, and the sound of MP4, MOV and MKV. Stereo is mixed to mono. Everything happens on your computer.
Meetings: Google Calendar#
Open the Meetings tab and press Connect Google Calendar.
- Open settings opens Google Calendar's settings in your browser.
- Under Settings for my calendars, pick your calendar, then Integrate calendar.
- Copy Secret address in iCal format.
- Come back to Hush. It notices the address on the clipboard, checks it with Google and shows the calendar's name and the meetings it found. Press Connect.
You can also paste the address into the field. Any calendar that publishes an iCal
(.ics) address over HTTPS works the same way.
| Setting | What it does |
|---|---|
| Check for changes every 5, 15, 30 or 60 min | How often Hush reads the calendar |
| Turn on noise cancellation 1, 2 or 5 min before | For meetings with a Meet, Zoom, Teams, Webex, Slack or Discord link |
| Turn it off after the call | Only if Hush turned it on, and you did not turn it off yourself |
| Pop up with a Join button | The tray widget opens a minute before the call |
Show one calendar#
With more than one calendar, the chips above Up next show All or one calendar's meetings. Hush remembers your choice.
More than one calendar#
Press + Add a calendar on the Meetings tab and connect another secret address, up to 12. The Calendars card lists each with its colour, the meetings it has, and whether it is syncing or needs attention.
- The switch pauses a calendar without forgetting it.
- ✕, then Remove forgets it.
- Coloured dots on a meeting show which calendars it comes from.
- The same meeting in two calendars (an invite on your work and personal calendar) shows once, and Hush pops up once.
If one calendar fails (its address was reset), the others keep working, and its row says what to do.
Join opens the meeting's own link. Remove makes Hush forget an address. The address is a secret link to your calendar: Hush keeps it only in its settings folder and never writes it to logs. If it leaks, reset it in Google Calendar and connect again.
Talk: Live Translate both ways (Preview)#
Talk with anyone in their language. They hear you in theirs through Hush AI Microphone; you read them in yours, and can hear them translated too. Hush uses Google's Gemini 3.5 Live Translate model with your own API key.
- On the Talk tab, press Add your key. Hush opens Google AI Studio.
- Press Create API key and copy it. Back in Hush, it is noticed, checked with Google and kept in Hush's settings folder.
- Choose I speak and They speak (70+ languages). ⇄ swaps them. Hush remembers them.
- Choose what to translate:
- Both ways: they hear you in their language, you read them in yours.
- My voice: only your voice is translated.
- Their voice: only theirs.
- Turn on Live Translate. Hush asks Start translating now? first, because voices go to Google. Tick Don't ask again to skip it, or turn it back on in Settings.
The Talk tab shows the call as a chat: their words in your language (what they said in small print), yours with what they heard. During a call, use the notch, the tray menu (Translate my voice, Translate their voice) or Ctrl + Alt + T (⌘ ⌥ T on macOS) in any app; the shortcut turns on the directions you chose.
With or without noise cancellation. Your voice does not need noise cancellation. With it on, Google gets your cleaned voice. With it off, Hush runs a quiet pass-through session.
Their voice: Hush AI Speaker#
Press Their voice on the Talk tab (or in Settings):
| Option | What it does |
|---|---|
| Clean their voice: Off, Light, Full | Removes their background noise before you hear it. Light keeps a little room sound. |
| You hear: Their voice | Their own voice; the translation shows as captions. |
| You hear: Voice-over | Their voice turns down while the translation speaks over it, like an interpreter on TV. The default: you hear who is talking and follow what they say. |
| You hear: Translation | Only the translated voice. |
| Play their voice on | Your real speakers or headphones. |
How Hush hears them:
- macOS and Linux: choose Hush AI Speaker as the speaker in your meeting app.
- Windows: with no setup, Hush listens along to your speakers and translates them (it cannot clean or replace their voice this way). To clean it too, install VB-CABLE A+B from VB-Audio and choose CABLE-A Input as the meeting's speaker; Hush finds it on its own.
When Hush quits, apps that played to Hush AI Speaker fall back to your speakers (Linux) or stay silent until Hush runs again (macOS, Windows cable): choose your speakers again in the meeting app if you stop using Hush.
Conversations#
Every call is kept as a conversation: one calendar meeting, or one stretch of talk (a pause of 10 minutes starts a new one). Back-to-back meetings stay apart.
- Search finds words in either language, meeting names and titles.
- Filters: All, Starred, Meetings, and a language.
- Rename with a click on the title. Star the ones that matter; starred ones are kept longest.
- Export one conversation, or all of them, as Markdown, text or JSON.
- Conversations stay on this computer (
conversations.jsonin Hush's data folder). Turn off Keep conversations in Settings and Hush stops saving new ones.
Good to know:
- It uses the internet. While a direction is on, that voice goes to Google; everything else in Hush stays on your computer.
- It follows by a phrase or so. Gemini translates phrase by phrase.
- Cost: Google offers a free tier; paid use is billed by Google to your key. Google's terms apply to the audio you send.
The notch#
Hush sits at the top of your screen as a small island (on a MacBook, it hides in the camera notch).
- Idle: a sliver. It costs nothing to keep.
- In a call (noise cancellation on and a meeting, their voice playing, or Live Translate on): it shows your level and what is happening, with a softly turning rim.
- Point at it (or click the tray icon) and it opens: noise cancellation, Live Translate, their voice, both sides' captions and your next meeting with Join.
Clicks pass through everywhere except the island. Prefer the tray widget? Settings › Quick controls › Show Hush as › Tray widget. Hide from screen sharing keeps Hush's windows and the notch out of screenshots, screen sharing and recordings (Windows and macOS).
Fix problems#
| Problem | Fix |
|---|---|
| Apps hear nothing | Check that Hush is On and the app uses Hush AI Microphone. |
| My voice sounds robotic or cut | Choose Efficient, or Light with RE-USE. Close heavy programs. On Windows, keep Audio priority on. |
| "Your microphone is not available" | Reconnect it or choose another. Hush retries on its own. |
| Windows: Hush AI Microphone missing after a VB-CABLE update | Open Hush › Set up › Rename again. |
| macOS: Hush AI Microphone missing | Settings › Hush AI Microphone, or run setup again (it restarts Core Audio). |
| Delay is noticeable | Windows with VB-CABLE adds about 84 ms. In VB-CABLE's control panel, a smaller buffer lowers it. |
| Studio: "Hush cannot read this file" | Convert it to WAV or MP3. Opus files (for example WhatsApp voice notes) are not supported yet. |
| Meetings: "The calendar did not accept this address" | The secret address was reset. Copy it again from Google Calendar. |
| Live Translate: "Google did not accept this API key" | Create a new key in Google AI Studio and add it again. |
| Live Translate is on but they hear silence | Check that the call app uses Hush AI Microphone, speak a full sentence and wait a moment. Gemini answers phrase by phrase. |
| Meetings: one calendar in the list shows an error | Its secret address was reset. Remove it and add the new address; the other calendars keep working. |
| Ctrl + Alt + T does nothing | Another app uses the same shortcut. Use the switch in the notch or widget instead. |
| I hear nothing from the meeting | The meeting app plays to Hush AI Speaker: keep Hush running, or choose your speakers in the meeting app. Check Play their voice on under Their voice. |
| Windows: Their voice says "Listening along" | No second cable is installed: translation works, cleaning their voice needs VB-CABLE A+B. |
Uninstall#
- Settings › Hush AI Microphone › Remove (macOS) or Restore VB-CABLE names (Windows).
- Uninstall the app like any other app. On Windows, uninstall VB-CABLE too if you no longer need it.
- Optional: delete the models and settings folder:
- Windows:
%APPDATA%\ai.hush.desktop - macOS:
~/Library/Application Support/ai.hush.desktop - Linux:
~/.local/share/ai.hush.desktopand~/.config/ai.hush.desktop
- Windows:
Privacy#
Noise cancellation and Studio run only on your computer. Hush connects to the internet to download a model (Hugging Face, or the model's original source as a fallback), to check for updates (GitHub), to read your calendar's iCal address if you connect one (Google), and to send voices to Google while Live Translate is on (yours, theirs, or both, as you chose). It sends no usage data and no crash reports, and it has no servers of its own.
Something unclear or out of date? Tell us or edit this page.