Lojban Speech Synthesis

From Lojban
Jump to navigation Jump to search

xVASynth

While the Open Source project supports only English with upcoming IPA support. A Lojban voice can be trained by replacing the ARPAbet dictionary with Lojban equivalent pronunciations. The voice of la selpa'i was synthesized using Lojban Corpus Readings as audio sources. The English ARPABet dictionaries have to be disabled for it to work properly.

File:xvasynth selpahi example 1.wav

xVASynth voice model: https://huggingface.co/Pendrokar/xvasynth_lojban/tree/main

Online Demo of the voice model, see the extra voice options: https://huggingface.co/spaces/Pendrokar/xVASynth

Selpahi dataset: https://drive.google.com/drive/folders/1ACqRYT7qf4rfSpnjIIbpqmoCTryMmF9w

eSpeak

The Open Source eSpeak supports Lojban.

Web Version

https://eeejay.github.io/espeak/emscripten/espeak.html

Select the "lojban (jbo)" voice.

Command Line Version

https://github.com/espeak-ng/espeak-ng

espeak-ng -v jbo

You can then type any Lojban text and press return for it to be read out-loud. Don't press any arrow keys when entering text, as that will cause problems - backspace works fine.

GUI Version

http://espeak.sourceforge.net/

Type in 'jbo' when installing the languages

Other software

http://members.home.nl/w.dijkhuis/lojban/z0300_txt2speech_01.htm has some more TTS examples.

There is also lojban_diphone_speech_synthesizer which uses Praat, but a lot of its links are dead now.