Reproduce “Bolbosh: Script-Aware Flow Matching for Kashmiri Text-to-Speech” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tajamul Ashraf, Burhaan Rasheed Zargar, Saeed Abdul Muizz, Ifrah Mushtaq, Nazima Mehdi, Iqra Altaf Gillani, Aadil Amin Kak, Janibul Bashir
Bolbosh is an open-source neural text-to-speech system for Kashmiri, adapted from pretrained English Matcha-TTS. It preserves pronunciation-critical diacritics in a 272-symbol vocabulary, fine-tunes on Kashmiri recordings, and generates speech using flow matching. It scored MOS 3.634 versus 1.864 for IndicParler and MCD 3.73 versus 4.73, remaining below human-reference MOS without isolating component effects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.