Open-source models, research projects, and professional work by Vadzim Belski.
Production-grade TTS models for Emirati Arabic dialect — bilingual FastPitch/HiFi-GAN, VITS end-to-end, and Qwen3.5-based TTS with extended text normalization pipelines.
Qwen3-based TTS model fine-tuned for Saudi Arabian Arabic dialect with dialect-specific text normalization and production-grade speech synthesis.
Real-time PCM streaming TTS with ~6x inference speedup and Arabic language support built on Qwen3-TTS. OpenAI-compatible API endpoint with Server-Sent Events streaming.
Fork of NVIDIA NeMo adding Emirati Arabic dialect support for VITS TTS. Custom EmiratiG2P module with IPA phonological rules for Gulf Arabic dialect-specific transformations.
Fork of NVIDIA NeMo text normalization adding Emirati Arabic (ar_ae) dialect support. WFST-based normalization for numbers, dates, currencies, and Gulf Arabic-specific entities.
AI-powered test automation agent using Claude AI SDK and Playwright. Supply a requirements doc and URL — the agent logs in, navigates the app, and generates verification reports.
Head of AI & Principal Architect at ScienceSoft since 2007 — scaled the AI/ML team, opened new expertise in Blockchain, AI, ML, and maintained AWS partnership.