Repository navigation
Home
Arachne compares a performance of a speech with a reference reading of the exact same text, measures where and how the delivery differs, and explains each difference with the feature, comparison, time span and rubric rule behind it.
A flagged region is a measurement against one chosen reference, tied to an explicit rule. It is not proof of a "wrong" delivery, of intent, or of any cause.
| I want to... | Go to |
|---|---|
| Run it | Getting Started |
| Understand the design | Architecture |
| Build the dashboard against the API | API Reference · Result Schema |
| See what each rule measures | Rubric |
| Understand the data | Dataset |
| Check how well it works | Evaluation and Results · Stress Testing |
| See what's done and what's next | Project Status · Roadmap |
| Know the caveats | Limitations |
| Work on it with the team | Team and Workflow · Contributing |
| Submit to Devpost | Submission Checklist |
| Quick answers | FAQ |
| Area | State |
|---|---|
| Backend pipeline (alignment, features, 4 rubric rules, scoring) | Complete, 9/9 tests passing |
| REST API and result schema | Complete |
| Synthetic dev dataset (63 items) + evaluation + stress harness | Complete |
| Dashboard | In progress |
| Real recordings and human-reviewed labels | Not started |
| WhisperX forced alignment on real audio | Implemented, untested |
| Technical document, demo video, Devpost | Not started |
Submission: internal deadline 14 Oct 2026; the event page lists 12:15 AM IST on 15 Oct 2026. Feature freeze 12 Oct.