Download
Stable · 334 commits since 26.03

26.10

Dubbing with a voice per speaker, engines as add-ons, the subtitld.cc hub inside the app, a vertical timeline, a plain-text view, seventeen interface languages, new subtitle formats, and Subtitld on macOS and Haiku.

Download
Subtitld 26.10 with the Speakers panel open and each speaker on their own timeline lane

Highlights

Dubbing

Speak every subtitle with a synthetic voice, one per speaker, fitted to each subtitle's duration. Voices that clone take their sample from the video. Export the mix, a track per speaker, the clips, or the background alone.

Record mode

Speak into the microphone and watch it become subtitles. Each phrase turns into a real subtitle at once, and its text fills in a moment later.

Add-ons

Speech recognition, translation, voices and audio separation install from Settings › Add-ons, each with its own models. Many run fully offline.

subtitld.cc

A panel to find, open, rate and publish subtitles on the community hub. A new subtitle is in the editor as soon as it is published.

The new vertical timeline tab beside the video
Vertical timeline. The timeline on its side, as a left-panel tab.
The new plain-text tab with the subtitles as SRT
Plain-text view. Edit all the subtitles as SRT, Markdown or JSON.

Before you upgrade

  • whisper.cpp is no longer built in. Install the whisper.cpp add-on to keep transcribing offline.
  • The built-in Google translation engine is gone. The Offline translation add-on replaces it, with no API key.
  • AssemblyAI moved from built in to a cloud add-on.
  • JSON is saved in Whisper's format, or as Bilibili's BCC. Opening a JSON file reads what other tools write.

New

  • DubbingSynthetic voices per speaker, a recorded voice as an alternative, and "Fit dub to subtitle duration" kept per speaker in the project.
  • AudioAn effects mixer with EQ, compressor and gate per track, and an input monitor for recording.
  • TimelineA lane per speaker, and the vertical timeline as a left-panel tab.
  • Plain textEdit the subtitles as SRT, Markdown or JSON; the text follows the selection, with font and colour theme on one row.
  • subtitld.ccFind, open, rate, publish and delete subtitles, with the client bundled and an update notice.
  • LanguagesFifteen more interface languages, seventeen in all. The system's language by default, or one picked in the settings.
  • VideoVideo properties with Source and Proxy tabs; 4K and 8K videos get a lighter proxy to edit with.
  • Projects.usfx files have an icon of their own and a preview: a card with the frame where the project was left, its subtitle and the timeline there. Double-clicked in a file manager, a project opens in Subtitld.
  • IconA new icon and logo.

Formats

Subtitld now reads and writes every format itself.

  • NewEBU STL, MacCaption (MCC), CEA-708, YouTube's timed text (srv3) and Bilibili's BCC.
  • JSONOpens what speech-to-text tools and services write: Whisper, WhisperX, Deepgram, AWS, Rev, Google, Azure, Speechmatics and more, split into subtitles where the speaker changes and where speech pauses.
  • From the videoIts closed captions (CEA-608 and CEA-708) and its own subtitle tracks.
  • RewrittenSRT, WebVTT, ASS and SSA, TTML and DFXP, SAMI, SCC, SBV, MicroDVD and USF: read as other tools write them, written as players show them.

Improved

  • Playback: picture and sound in sync, timed from the sound device itself (through WASAPI on Windows).
  • Undo: a whole drag, or a held nudge key, is one step.
  • Subtitle edges: the glued neighbour follows, frames snap, and 0.0 counts as a time.
  • Speakers: face detection on current MediaPipe, through the Tasks API; seek from a speaker's strip.
  • Find & Replace steps both ways from the cursor.
  • Autosave keeps backups grouped by day.
  • Keyboard shortcuts panel: a command list and a capture area.
  • Transcription, translation and dubbing share one footer, and check what they need before they start.
  • Selecting a subtitle in the list no longer starts editing it.

Fixed

  • The timeline follows the playhead again while playing.
  • The timeline's empty-state label no longer crashes the paint.
  • Audio starts without an output device instead of aborting.
  • The Add-ons panel finishes its catalog fetch and installs on quit.
  • Translation retries errors instead of storing the error page as text.

Packaging

System What changed
macOS New: the app for Apple Silicon in a .dmg. It is not signed: the first time, allow it in System Settings › Privacy & Security.
Haiku A native application with its icon, and .usfx a file type of its own; HaikuPorts recipes built and started in a Haiku VM.
Windows A Modern UI installer, for you or for everyone, in two languages, with its own icons; also an MSIX package and a portable .zip. Project previews show in Explorer.
Snap From 650 MB to 424 MB, by leaving out what the app never uses.
Flatpak Back on Flathub, with the KDE runtime and PySide 6.11, every Python dependency, and store screenshots.
AppImage Uses the bundled PortAudio when the system has none.
AUR Complete dependencies, built the way an Arch user would, and project previews in the file manager.
PyPI Published with each release, as a wheel and a source archive.

Removed

  • The built-in whisper.cpp, Google translation and AssemblyAI engines, now add-ons.
  • The pycaption, deep-translator, translate, psutil and autohex dependencies.