LiveSub v1.3.0

Clearer speech in noisy audio, and subtitles that come out translated.

New

• Background noise removal before speech is recognized (Settings → Recognition): Normal by default, Strong for loud backgrounds, or Off. It removes crowd, street and other background sound so the voice is heard. The first Start downloads its small model (0.5 MB).

Fixed

• Subtitles sometimes stayed in the original language. On short, noisy pieces the speech model could mislabel Korean or Japanese as English, and the whole line was shown untranslated. Each sentence now uses the language most of it was heard in, checked against its alphabet.
• Mixed-language lines (English after Korean, for example) are now separate sentences, so each is translated properly.
• Korean words were glued together when pieces were joined. They now keep their spaces.
• Japanese sentences cut in half after words like まだ, ただ or 誰か.
• Repeated words in long non-stop speech (About 1 second mode).
• Several sentences in one line with Instant models that add punctuation (Kroko, X-ASR).
• The last words before Stop are now always recognized.
• No "pack missing" warning when the spoken and subtitle language are the same.

Install

Install LiveSub-v1.3.0.apk below (Android 10 or newer). It updates v1.2.1 in place and keeps your models and history. Coming from v1.2.0 or older? Uninstall that first (the package name changed in v1.2.1).

This build is signed with a debug key, so Android may warn that it's from an unknown developer.

Built from main (commit 3c961db).

LiveSub v1.2.1

LiveSub has a new package name, ready for the Play Store. The app itself is the same as v1.2.0.

Important: uninstall the old version first

From this version the app's package name is io.github.rdilipkumar.livesub (it was com.aistudio.liveaudiotranslate.zqxkpv). Android treats it as a different app, so it won't update v1.2.0 or older; it installs next to it.

1. Uninstall the old LiveSub.
2. Install LiveSub-v1.2.1.apk below (Android 10 or newer) and allow installing from this source when asked.
3. Download your speech models and translation packs again in the Models tab. History from the old app doesn't carry over.

Future versions install over this one as normal updates.

This build is signed with a debug key, so Android may warn that it's from an unknown developer.

What's in it

Everything from v1.2.0:

• sentence-by-sentence subtitles;
• all 59 ML Kit languages (including Tamil and other Indian languages);
• up to 8 lines in the floating window, with a scroll-back limit;
• crash fixes.

Built from main (commit 43941ee).

LiveSub v1.2.0

Subtitles now arrive sentence by sentence, every language ML Kit can translate is available, and the floating subtitles can show more lines. Installs over v1.1.0 and v1.0.0 (same signing key); your models, packs and history are kept.

Install

1. Download LiveSub-v1.2.0.apk below on your Android phone (Android 10 or newer).
2. Open it and allow installing from this source when asked.

This build is signed with a debug key, so Android may warn that it's from an unknown developer.

What's new

Subtitles sooner, sentence by sentence ("About 1 second" models: SenseVoice, Whisper, Moonshine, Parakeet, GigaAM)

• No more waiting for long pauses: speech is sent at sentence breaks, and recognized in the background while the next part is still being heard.
• It adapts by itself:
  • To the speaker: it learns their rhythm, so fast talkers' short breaks are caught.
  • To your phone: it learns how fast the model runs, so subtitles keep up.
  • In long non-stop speech: it cuts between words, never in the middle of one.
• Each sentence gets its own subtitle line or card, and its own history entry.
• A translated draft shows right away; the full sentence replaces it when it's done.
• Whisper runs about twice as fast (it now uses 4 cores). If a model is still too slow for your phone, the app tells you and suggests a faster one.
• Whisper's sound labels like "[Music]" or "laughs" no longer show up as subtitles.

59 languages

• The spoken-language and subtitle-language lists now include every language ML Kit can translate. That includes Tamil, Telugu, Kannada, Marathi, Bengali, Gujarati and Urdu.
• For speech in Indian languages, use a Whisper model (Whisper Base is the best fit for most phones).

Bigger floating subtitles

• Up to 8 lines: drag the corner (up or down for lines, sideways for width), or use the new "Lines" control in the window's settings.
• New "Scroll back" limit: choose how many past subtitles you can scroll back to (10–200), in the window's settings or in Settings.
• The window moves back on screen if it grows past the bottom.

More reliable

• Fixed several ways the app could close unexpectedly: when stopping, if recognition hit an error, or when Whisper was set to a language it doesn't know (like Irish or Esperanto).

Built from main (commit f6ffc0f).

LiveSub v1.1.0

Big accuracy and subtitle update. Installs over v1.0.0 (same signing key); your models, packs and history are kept.

Install

1. Download LiveSub-v1.1.0.apk below on your Android phone (Android 10 or newer).
2. Open it and allow installing from this source when asked.

This build is signed with a debug key, so Android may warn that it's from an unknown developer.

What's new

Choose speed or accuracy (Settings → Preferred latency)

• Instant: words appear while they are spoken (streaming models, as before).
• About 1 second: text appears when the speaker pauses, with punctuation and far fewer mistakes. Each latency keeps its own model.
• New whole-phrase models: SenseVoice (Japanese, Chinese, English, Korean, Cantonese; best for anime and dramas), Moonshine (8 languages), Parakeet v3 (25 European languages), GigaAM v3 (Russian) and Whisper Base/Small (99 languages).
• SenseVoice and Whisper can be locked to the spoken language, so Japanese is no longer mistaken for Chinese.

Better subtitles

• New Roll-up style (default): lines stack and roll up; scroll back to reread, and it returns to live by itself. Movie and Live styles are still there.
• Whole sentences are translated together, so meaning isn't cut in half.
• Music, laughter and noise are no longer turned into random text.
• The floating window opens when you press Start, can be resized, and hides its controls on double-tap.

Clearer app

• New main screen: shows which speech model is listening and whether it understands your language.
• The Models screen marks the model in use for each latency; downloading a model no longer switches to it.
• A banner tells you when a translation pack is missing, instead of showing nothing.

Fixes

• Fixed crashes when a model download was interrupted, and with Moonshine and Whisper Japanese.
• The last sentence before Stop is now saved to history.
• Stopping from the notification or the Quick Settings tile also closes the floating subtitles.
• The Quick Settings tile asks you to allow audio recording first instead of failing.
• Smoother overlay settings slider; more reliable language pack download progress.
• Recorded debug audio is never included in backups.
• Removed the "No app audio / DRM" warning from the overlay.
• Removed unused cloud libraries (Firebase, Retrofit, Moshi); models only download over HTTPS.

Built from main (commit bf29b0c).

LiveSub v1.0.0

LiveSub turns speech from any app on your phone into live, translated subtitles. Recognition and translation run on the phone.

Install

1. Download LiveSub-v1.0.0.apk below on your Android phone (Android 10 or newer).
2. Open it and allow installing from this source when asked.
3. In the app, open Manage Models, download a speech model (Multilingual for Japanese) and the translation packs for the language you listen to and the one you read.
4. On the Translate tab, press Start, then Launch Overlay to show subtitles over other apps.

This build is signed with a debug key, so Android may warn that it's from an unknown developer.

Highlights

• Live speech recognition with one streaming model per language: Multilingual (Japanese, Chinese, English, Russian, Arabic, Thai, Vietnamese, Indonesian), English, Chinese, Spanish, French, German, Korean, Russian and Bengali.
• On-device translation with automatic language detection.
• Floating subtitles that always show the newest words; live feed and history in the app.
• Audio tuned for speech, including quiet or husky voices.
• Simple Settings, plus Developer settings for recognition tuning and troubleshooting.

Built from main (commit 0428222).

LiveSub v0.3.0 beta

Test build of LiveSub, the on-device live subtitle translator.

Install

1. Download LiveSub-v0.3.0-beta.apk below on your Android phone (Android 10 or newer).
2. Open it and allow installing from this source when asked.
3. In the app, open Manage Models, download a speech model (Multilingual for Japanese) and the translation packs for the language you listen to and the one you read.

This build is signed with a debug key, so it can't be installed over a Play Store version, and Android may warn that it's from an unknown developer.

What's new

• Fresh recognition stream after each sentence, so recognition no longer stops after a few sentences.
• Cleaner audio path: proper 16 kHz resampling, full speech band, quiet-voice boost; noise gate off by default.
• One streaming model per language (Multilingual, English, Chinese, Spanish, French, German, Korean, Russian, Bengali), with real sizes.
• Ordered translation, 5 s timeout, auto-detect language with ML Kit.
• Overlay always shows the newest words; live feed shows newest at the bottom.
• Simple Settings plus Developer settings for recognition tuning (pause length, restart after no words, shortest sentence, one-character lines) and debug recording.

Built from main (commit 0428222).