Spaces:
Sleeping
Sleeping
File size: 1,373 Bytes
51a27d1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 | # Scope And Success Criteria
## Current Supported Vocabulary
The current word-level target set remains intentionally small and demo-friendly:
- `HELLO`
- `YES`
- `NO`
- `THANK_YOU`
- `PLEASE`
- `HELP`
- `STOP`
- `EAT`
- `DRINK`
- `WANT`
- `MORE`
- `FINISHED`
## Current Baseline Success Criteria
The current baseline is successful when:
- a teammate can clone the repo and install it locally
- `run_demo` produces at least one translated event and transcript entry
- `train_model` saves a baseline artifact from the sample dataset
- `evaluate_model` produces metrics without manual editing
- the output path stays resilient even if ElevenLabs, VLM access, or webcam access is unavailable
## Next Scope: Sentence Mode
The next project target is not full open-ended ASL translation. It is controlled sentence-level interpretation over short gesture windows.
Sentence mode should support:
- short 2-5 second clips
- sampled keyframes from the clip
- a token trace from the existing word-level classifier
- VLM interpretation into one English sentence
- mock/offline fallback when the VLM is unavailable
## Still Out Of Scope
- open-ended full ASL translation
- long continuous conversations
- production accessibility claims
- production deployment
- mandatory Firebase integration
- mandatory cloud TTS during local demos
- mandatory cloud VLM during local tests
|