Logo Lanfrica

Shark-Fujin/Bridgelingua

Domaine:

natural language processing

Type de record:

softwaretools
Créateur:
Sha
Hôte:
Open-source web platform for multilingual speech recognition across 1,600+ languages. Transcribe, translate, and archive endangered language audio with AI (Meta OmniASR/MMS + NLLB). Community-driven fine-tuning for low-resource languages. Built for field linguists & language preservation / UNESCO IDIL. # Bridgelingua **Open-source web platform for multilingual speech recognition across 1,600+ languages.** Transcribe, translate, and archive endangered language audio with AI. Built for field linguists & language preservation communities. A UNESCO IDIL 2022-2032 registered initiative. --- ## What is Bridgelingua? Over 40% of the world's 7,000+ languages are endangered — most have never been written down. Recent AI breakthroughs (Meta's OmniASR/MMS) now support ASR for 1,600+ languages, but these models remain locked in research environments inaccessible to the people who need them most. **Bridgelingua bridges this gap.** We provide field linguists, indigenous communities, and language preservation practitioners with an accessible web platform to document, transcribe, translate, and archive any language — no coding required. ## Features - **1,600+ Language ASR** — Automatic speech recognition powered by Meta OmniASR/MMS, covering more languages than any commercial service - **Bilingual Translation** — Neural machine translation (NLLB) pipeline connecting indigenous languages to global languages - **Multi-format Export** — JSON, CSV, SRT, TXT, with ELAN XML support planned - **Community Fine-tuning** *(P2)* — Upload labeled speech data to fine-tune ASR models for your language via LoRA - **Ethical Data Governance** — Built with CARE Principles for Indigenous Data Governance - **i18n Interface** — Full Chinese/English bilingual UI ## Tech Stack | Layer | Technology | |-------|-----------| | Frontend | React 19 + TypeScript + Vite + Zustand | | Backend | Python 3.10+ / FastAPI + SQLAlchemy (async) | | ASR Model | Meta OmniASR (Wav2Vec2 + LLM, 1,600+ languages) | | Translation | Meta NLLB-200 (200 languages) | | Database | SQLite (async via aiosqlite) | | Deployment | Nginx + systemd (Linux) | ## Quick Start ### Prerequisites - Python 3.10+ - Node.js 18+ - GPU with CUDA 11.8+ *(optional — runs in Mock mode without GPU)* ### Backend ```bash cd backend python …