Voice Subtitles
Live subtitles for Simple Voice Chat: offline speech recognition turns what people say into captions, and translates them when they are speaking another language. All of it on your own machine.
Список изменений
Two things, both found on a live server.
Everything after the first sentence was lost
Marian, which is the shape every OPUS-MT model has, translates one sentence and then stops, whatever it is handed. "Hey my name is Denis. What is your name?" came back as "Меня зовут Денис." and nothing else. The same two sentences the other way round came back as the other half, which read as though the model were shuffling them rather than dropping them.
A line is now cut into sentences and each one goes through on its own. Cut on a full stop, a question mark or an exclamation mark followed by a space, and nowhere else: not after an initial, not after a bare number, not inside 3.14 or example.com, and a closing quotation mark stays with the sentence it closes. A line with one sentence in it, which is nearly all of them, costs exactly what it did before.
The language list read "en, en, en, en"
What /vsub lang <player> offered was built from the recognition catalogue, which holds four English models and nine Russian ones. So it offered en four times and ru nine, every line identical, with the only thing telling them apart hidden in a tooltip.
It is now the two languages there is a translation for, offered once each, with the name of the language beside the code. And the name is used wherever the code used to be printed: in the answer to /vsub lang, in the line that says what everybody here is taken to speak, and in the list of who has been marked as speaking something else.
Two is all there is, because two is what this mod carries: en-ru and ru-en, a hundred and thirteen megabytes each, and every further pair would be another one of those. A language that is only in the recognition catalogue – German, French, Japanese – is still accepted when it is typed out, since a caption in the right alphabet is worth having even where nothing can translate it. It is simply not suggested.
Two files, as before
voice-subtitles-1.2.2.jar is the ordinary one: 3 MB, and it fetches the engine and the model the first time it needs them. Take this one.
voice-subtitles-1.2.2-full.jar is 177 MB and is the same mod with the recognition engine and the Russian GigaAM model already inside it. On the first run it copies them onto disk in about a second and never asks the network for anything. It is for people whose problem is the download rather than the size; the engine in it is built for Windows x64.