GrammaScore is a machine learning-based Grammar Scoring Engine designed to evaluate the grammatical quality of spoken audio clips. It predicts a continuous score between 0 and 5 using MOS (Mean Opinion Score) Likert-scale ratings.
In this project, we build a regression model that listens to audio samples (45–60 seconds each) and predicts a grammar quality score. This can be applied to spoken language assessment, pronunciation tools, or AI-based interview evaluators.
- Input:
.wavaudio file - Output: Grammar MOS Score ∈ [0.0, 5.0]
- Training samples: 444 labeled audio clips
- Test samples: 195 unlabeled audio clips
Each clip is annotated with a grammar proficiency score based on a Likert-scale MOS rubric.