VoiceLK, Sinhala-English speech synthesis

Field notes

The page describes a pipeline that first expands numbers, dates, abbreviations, and mixed scripts into spoken forms, then decides language word by word and maps the output to a shared phoneme sequence. A VITS model converts that sequence to a waveform; the site says it tracks multi-speaker and single-speaker training checkpoints and stores each rendered clip with its producing checkpoint. A small FastAPI bridge is named as the serving layer.