Repository navigation
Releases
v0.4.0
Highlights
- Live translation. Pick one language in settings ("Translation") to translate everything into it, or two to translate between them. Translated lines appear in italics under the speech they translate and are saved in the transcript. For one run:
earpop --translate koorearpop --translate en,es. - Built for mixed-language conversations. Every line shows its detected language when more than one is in play, such as
[S1 ja], and a new line starts when the language changes. Presslto show one language at a time. - Captions that read better in every script. Lines now end at sentence boundaries in Latin, CJK, Hindi, Arabic, and Urdu text, after a pause of more than two seconds, and when the speaker or language changes. Japanese, Chinese, and Thai wrap between words, and a speaker tag is never left alone on a row.
- Captions fill the terminal. A taller window shows more of the conversation, and settings, the microphone list, and the new
?key help open below the captions instead of hiding them. - Subtitle and Markdown exports.
earpop transcripts export <id> --format srt|vtt|md, with--outfor the path andlatestfor the newest transcript. Subtitle cues end at sentences, and translations appear at the same time as the speech. - Terms.
earpop terms add "Kold Brew" Kubernetesteaches Soniox how to spell names and jargon. - Languages on first run. earpop asks which languages people will speak instead of assuming the system language.
--lang ja,enoverrides them for one run, the language picker is searchable, andearpop languageslists every code. - Bluetooth microphones and hearing aids. If a microphone stops sending audio, as Bluetooth devices do when they switch to their microphone mode, earpop reopens it within a few seconds instead of showing "Listening" with no captions.
- Soniox stt-rt-v6. The newest model, with better accuracy across languages. stt-rt-v5 is retired by Soniox on 2026-12-15.
Changes
earpop transcripts listshows each transcript's length instead of its token count, and reads only the start and end of each file.- Pausing no longer copies the transcript path to the clipboard. Press
cto copy it. - The key hint line keeps the most important keys on narrow terminals;
?lists them all. Shortcuts also accept Korean jamo typed in Hangul mode. - The status line shows the languages in use and any translation.
- On Linux, the microphone list hides ALSA's converter and routing plugins and shows the default input as "System default". Microphones saved under the old name still work.
- Speaker colors no longer use white or dark blue, which disappeared on light or dark themes.
earpop --versionprints the version.- In our tests, provisional captions arrive about 0.3 seconds later with stt-rt-v6 than with v5.
Fixes
- Sentences ending in a period now start a new line; before, a line ran on until the speaker changed.
ESCworks on the region, API key, and first microphone screens; the key screen's ESC goes back to region selection.- A saved microphone that is no longer connected no longer hides the microphone list on first run.
- Settings and microphone menus fit short terminals.
- Losing the microphone closes the Soniox session until you pick another one, so no idle session stays open.
- The capture helper exits when earpop is killed, so the microphone is never left open.
- Errors that a reconnect cannot fix (bad request, key, balance, or permission) stop instead of retrying forever.
- A transcript that cannot be saved shows a warning on screen and at exit.
- Failed settings saves show a message instead of crashing.
Ctrl+Cno longer also triggers the copy shortcut.
Under the hood
- Sends the Soniox API key in the
Authorizationheader; the start-requestapi_keyfield is deprecated. - The published CLI runs React and Ink in production mode, which uses less memory on every caption update.
- Updated ws (includes a security fix), oxlint, oxfmt, and Rust crates; the
rules-of-hookslint rule is on. - Unit tests (
pnpm test) and a CI check for every pull request. - macOS capture helpers now build on
macos-15.
Install
npm i -g earpop-cli@0.4.0Platform capture packages (earpop-capture-*) are published as matching optional dependencies.
Full changelog: v0.3.0...v0.4.0
v0.3.0
Highlights
- Caption history. Press
Upor scroll the mouse wheel to read the whole session at full screen height.Up/Down,PgUp/PgDn, andHomemove through it;EndorESCreturns to live captions. History also works while paused and after the session ends. - Bottom-anchored live screen. Captions grow upward from the status and hint lines, so opening the history no longer shifts the screen.
- A status line that tells you what is going on. A lost microphone shows as an error with a hint to press
m, reconnects show the reason and a countdown, microphone and language changes show a short notice, and the current microphone is shown next to the clock. Pause and resume react immediately. - Safer quitting. Ending a session from the live screen takes two
ESCpresses within two seconds, so backing out of settings no longer ends it by accident. The next press quits once the transcript is saved, and one more quits right away.Ctrl+Cstill ends the session at once.
Fixes
- The last sentence before stopping, pausing, or changing languages is no longer lost. earpop now ends the audio stream the way Soniox expects, so quitting also takes about a second instead of five.
- Picking a microphone on first run no longer starts several transcription sessions at once.
- Token timestamps stay correct after switching microphones mid-session.
earpop transcripts viewandexportkeep the spaces between words.- A new session after a pause or language change keeps a space after the previous sentence (not added for scripts written without spaces).
- The current microphone can be picked again after it disconnects, and a failed switch keeps the working microphone.
- Quitting while the microphone is still opening cleans up the capture helper.
- If Soniox ends a session on its side, earpop reconnects instead of showing "Listening" with no captions.
- Keychain or settings load failures show an error screen instead of crashing.
- Caption wrapping is correct for emoji, Thai, Hindi, and other wide or combining characters.
- Resizing the terminal no longer leaves a blank screen while paused or ended.
- Pasted API keys are trimmed before they are verified.
- Deleting the saved API key now asks for confirmation with a standard y/N prompt.
Under the hood
- Upgraded to Ink 8, React 19.3, ws 8.22, tsup 9, and current oxlint and oxfmt; Rust crates updated.
earpop transcriptsandearpop auth clearstart faster because they no longer load the terminal UI.- Shared helpers replace duplicated code across the CLI and release scripts.
Install
npm i -g earpop-cli@0.3.0Platform capture packages (earpop-capture-*) are published as matching optional dependencies.
Full changelog: v0.2.3...v0.3.0
v0.2.3
Install
npm i -g earpop-cli@0.2.3Platform capture packages (earpop-capture-*) are published as matching optional dependencies.
What's Changed
Full Changelog: v0.2.2...v0.2.3
v0.2.2
Install
npm i -g earpop-cli@0.2.2Platform capture packages (earpop-capture-*) are published as matching optional dependencies.
What's Changed
Full Changelog: v0.2.1...v0.2.2
v0.2.1
Install
npm i -g earpop-cli@0.2.1Platform capture packages (earpop-capture-*) are published as matching optional dependencies.
What's Changed
- ci: 태그 푸시 시 npm publish 자동화 by @e2goon in #2
- feat: 언어 힌트 다중 선택과 라이브 설정 by @e2goon in #3
- feat: 라이브 화자 라벨·숫자키 포커스 by @e2goon in #4
Full Changelog: v0.2.0...v0.2.1
earpop-cli 0.2.0
earpop-cli 0.2.0
Live STT capture no longer depends on ffmpeg. Mic listing and streaming use a native cpal + rubato sidecar, shipped per platform via npm optional dependencies.
Highlights
- Native capture sidecar —
earpop-capture(cpal + rubato) streams 16 kHz mono PCM; no ffmpeg on the live path - Cross-platform packages — optional deps for darwin-arm64 / darwin-x64 / win32-x64 / linux-x64 / linux-arm64
- Integrity checks — SHA-256 verification of capture binaries (override with
EARPOP_CAPTURE_BINorEARPOP_SKIP_CAPTURE_INTEGRITY=1) - STT defaults — Soniox realtime with Korean/English language hints and diarization
- Node.js 24+ — engines and build target aligned to Active LTS
- UX — Esc quit shortcut fixed
Install
npm i -g earpop-cli@0.2.0
# or
npx earpop-cli@0.2.0npm installs only the capture binary that matches your OS/arch.
Requirements
- Node.js 24+
- Soniox API key (region must match CLI:
us/eu/jp) - Supported OS for live mic: macOS, Windows, Linux (ALSA on Linux)
Breaking / migration
- ffmpeg is no longer used for live capture. If a previous setup relied on a system ffmpeg binary, it can be removed for earpop.
- Node < 24 is unsupported.
- First install on a new platform pulls the matching
earpop-capture-*optional package; if optional deps are skipped (--omit=optional), live mic will not work until that package (or a local build) is available.
Platform packages
| Package | Platform |
|---|---|
earpop-capture-darwin-arm64 |
macOS Apple Silicon |
earpop-capture-darwin-x64 |
macOS Intel |
earpop-capture-win32-x64 |
Windows x64 |
earpop-capture-linux-x64 |
Linux x64 |
earpop-capture-linux-arm64 |
Linux arm64 |
Env (new / relevant)
| Variable | Role |
|---|---|
EARPOP_CAPTURE_BIN |
Path to a custom capture binary (skips integrity) |
EARPOP_SKIP_CAPTURE_INTEGRITY |
1 to skip SHA-256 checks |