Voice Subtitles

The mod is 3 MB instead of 27

Both recognition engines used to be carried inside the jar, built for Windows, Linux and macOS, which meant every player downloaded three platforms of an engine they would load one of. They are fetched when they are first needed now, for the platform in hand, and checked against a hash built into the mod before anything is unpacked or loaded.

Much better recognition

A second engine, sherpa-onnx, and nine models to choose from.

• ru-giga-rnnt is the new Russian default: GigaAM v3 by Sber, 162 MB, and it writes punctuation and capital letters of its own accord. The old default got roughly every third word wrong on spontaneous speech; this one is in another class entirely.
• en-fast is the new English default: NVIDIA Parakeet, 103 MB, punctuation included.
• en-heavy (465 MB) covers twenty-five European languages, Russian and Ukrainian among them, which is worth knowing before downloading two models for a bilingual server.
• ru-big-int8 is the accuracy of the old 1.8 GB Vosk model in 57 MB, and seconds to load rather than a minute.
• Japanese added.

The accurate models read a whole sentence before deciding what a word was, so a line arrives a second or so after it was said. The audio is re-read as it accumulates, so the caption still grows while somebody is talking.

A conversation log worth the name

• Kept across restarts. Written as you play, one file per day, held for a fortnight by default. Nothing leaves your machine.
• Search and filters: by text, by speaker, by voice or chat, by how far back, oldest or newest first. Matches are picked out in the lines.
• The rest of the evening too: joining a server or opening a world, deaths, commands as they were typed, the whole of the chat. Login and registration are never written down; they carry a password.
• Sealed. Each line is encrypted and carries the fingerprint of the one before it, so the file cannot be reworded in a text editor and any tampering is reported rather than shown as though genuine. It is not proof of anything to anybody, since the key sits on the same machine, but it turns silent tampering into visible tampering.
• Saving it writes what the filters are showing, and the line that says so opens the folder with the file picked out.

Captions that move

• Lines fade in and out over a length you can set, rather than appearing between one frame and the next.
• Recognised text is revealed a few characters at a time, so a caption grows the way speech does instead of arriving in jumps. It never delays anything: the reveal outruns the speaking.
• /vsub preview puts a sample caption on screen, so all of the above can be adjusted without waiting for somebody to talk.
• Captions above heads gained their own size, row limit, background and name toggle.
• An optional arrow in front of each line points at the speaker: up for ahead, down for behind you. Off by default.

While you wait

Downloading and installing both say so now, with what is being fetched, how much of how much, and roughly how long is left. Shown on menus as well as in game, since the first download starts at the title screen.

A third place to put subtitles

Voice and chat each gained log only: written down, drawn nowhere. Text chat is already readable in the chat window.

Alongside Absolute Cinema Camera

That mod hides the whole interface for its film mode, subtitles included. They stay now, sitting just on top of the bottom matte: somebody reading the conversation instead of hearing it should not lose it to a camera angle. The hotbar allowance is dropped while the hotbar is gone, so the captions sit where film subtitles sit rather than halfway up the picture. Nothing to install or switch on, and nothing changes if that mod is not there.

Fixed

• A caption could be filed twice, or half-written, when a long sentence took longer to decode than the feed took to forget it.
• The name was printed twice on servers that write <Name> words or [4] Name 1... [Fail].
• A long utterance was cut at twenty seconds, mid-word.
• Answers to your own /try and the like no longer become captions.
• The log's scrollbar can be dragged. It was being drawn outside the list it belonged to.