macOS · v1.2

Aloud

Menu-bar voice input. Tap Fn to start talking, tap again to stop, and the recognized text is injected straight into the focused field. The backend is Volcano/Doubao streaming ASR — it handles mixed Chinese-English and code terms, with an optional LLM pass that only fixes obvious mis-hears, never rewrites.

Download for macOS .dmg · Apple Silicon First time? Read the setup guide →
Tap Fn, talk — text lands in the focused field, live
Fn toggle

Hold Fn to talk and release to stop, or tap once to lock recording and tap again to stop. 90-second hard cap as a safety net.

Live text

A capsule overlay types text word-by-word as you speak — you see the recognition live, not after you've finished. Live waveform; on stop it injects and restores your full clipboard (images, files, rich text included).

Esc to cancel

Realize mid-recording you misspoke? Hit Esc to cancel outright, nothing gets injected, nothing is left behind, and your clipboard stays untouched.

Mixed lang

Volcano/Doubao streaming ASR 2.0, automatic Chinese-English code-switching, sharper on technical terms than system dictation.

LLM fixup

An optional Doubao seed-lite pass that fixes only obvious speech mis-recognition — no polishing, no rewriting. Can be turned off.

Local

Credentials stored on-machine, triggered locally, recognition goes straight to Volcano with no third-party relay.

Changelog

v1.2
2026-09-16
  • Hold-to-talk: hold Fn for more than half a second and release to stop. A quick tap still locks recording.
  • Trailing full stop is dropped in chat apps (WeChat, Feishu, Slack, Telegram and others).
  • Clipboard is only restored if you have not copied anything new in the meantime.
  • Very short phrases (10 characters or fewer) skip the LLM pass. Transient network errors retry once.
v1.1
2026-07-02
  • Esc cancels the current recording: nothing injected, nothing left behind.
  • Full clipboard restore (images, files and rich text survive).
  • Start / stop sound cues.
v1.0
2026-05-16
  • First public release: Fn toggle, Volcano/Doubao streaming ASR, optional LLM correction.

Requirements

  • macOS 14 Sonoma or later
  • Apple Silicon (M-series)
  • A Volcano Engine account — provision Doubao streaming ASR yourself, put AppID / Access Token into the app's settings
  • Microphone + Accessibility permission (required to monitor the Fn key and inject text)

This build is unsigned and unnotarized. On first launch macOS may say it's "damaged" — that's Gatekeeper blocking an unsigned download, not actual damage. Move Aloud to Applications, then run xattr -dr com.apple.quarantine /Applications/Aloud.app in Terminal and open it normally. It's an early tool I built for myself: no CI, no code signing, no auto-updater.