Loru

Loru is an offline sign language toolkit: landmark sequences → gloss/text, and sign → voice (WAV) — demos and train loops without a GPU for the smoke path.

Product: mergeos-bounties/Loru

---

Table of contents

• Highlights
• Desktop GUI (Qt)
• Screenshots
• Quick start
• CLI reference
• Data & pipeline
• Diagrams
• Repository layout
• Development
• MergeOS bounties
• License

---

Highlights

Mode	Description
Sign → text	Landmark JSON sequences → gloss / sentence
Sign → voice	Recognition + TTS-style WAV export
Offline demo	Samples, toy train, infer hello end-to-end
Desktop GUI	Modern PySide6 app (loru-gui)
Gloss vocab	Default gloss set for demos
Serve	Optional FastAPI for integrations

---

Desktop GUI (Qt)

Modern dark PySide6 demo shell — full demo, samples, infer, train, gloss vocab.

pip install -e ".[gui]"
loru-gui
# or: loru gui

Full offline demo (train → text → WAV)

Landmark sample catalog

Inference (sign→text / voice / gloss)

Toy train

Default gloss vocabulary

---

Screenshots

CLI / pipeline captures:

Pipeline	Samples
Offline sign → text → voice	Gloss sample catalog

---

Quick start

cd Loru
python -m venv .venv
.\.venv\Scripts\activate
pip install -e ".[dev,gui]"

loru version
loru data list
loru demo
loru-gui

Demo writes audio under the configured output directory (e.g. demo_hello.wav).

---

CLI reference

Command	Purpose
loru version	Version + demo gloss vocab
loru demo	Train smoke + infer text + voice on hello
loru gui / loru-gui	Qt desktop app (needs .[gui])
loru data list	Landmark sample files
loru samples list [--gloss TEXT]	Sample catalog with language and frame counts
loru infer demo -s hello	Gloss → sentence
loru infer text …	Sign file → text
loru train / eval	Toy train + evaluation
loru serve	Optional API

loru infer demo --sign thank_you
loru demo
loru-gui

---

Data & pipeline

samples (JSON landmarks)
        │
        ▼
  toy train / vocab
        │
        ├─► sign_to_text  → gloss / sentence
        └─► sign_to_voice → WAV path

Path	Content
Samples	SAMPLES_DIR landmark sequences
Outputs	OUT_DIR audio + reports

Respect consent and privacy for real sign recordings; demos use synthetic/offline fixtures.

---

Diagrams

System architecture and workflow — full width. Open the HTML files for dark/light theme and export (PNG/SVG).

Architecture

Open interactive diagram

Workflow

Open interactive diagram

Generated with archify.

---

Repository layout

src/loru/
  cli.py
  gui/            # PySide6 desktop demo (loru-gui)
  infer/          # text, voice, pipeline
  data/loader.py
  models/vocab.py
  train/toy_train.py
docs/screenshots/
docs/diagrams/

---

Development

pytest -q
ruff check src tests
loru demo
python scripts/capture_gui_shots.py   # refresh GUI screenshots

---

MergeOS bounties

High demand: sign packs (gloss + evidence photo/video + consent).
Star repos → claim issue → PR to master → MRG 25–200.

---

License

MIT · MergeOS / ThanhTrucSolutions