SHL- Grammar Scoring Engine for Voice Samples

🎤 Predict Grammar Scores from Spoken Audio

🧠 Objective

Build a machine learning model that can automatically evaluate spoken audio and assign a grammar score (1–5) based on sentence structure and syntax quality.

---

🗂️ Dataset Overview

Mind Map - Dataset
├── Audio Files (.wav)
│   ├── audios_train/
│   └── audios_test/
├── train.csv
│   └── filename + grammar score
├── test.csv
│   └── filename only
└── sample_submission.csv
    └── sample format for output

---

## ⚙️ Workflow / Pipeline

Mind Map - Workflow

1. 🎧 Audio to Text └── Using Whisper for transcription
2. ✨ Text Cleaning └── Remove punctuation, lowercase, clean spaces
3. 🧮 Feature Extraction └── TF-IDF Vectorizer (max 1000 features)
4. 🌲 Model Training └── Random Forest Regressor
5. 📊 Evaluation └── Pearson Correlation
6. 🧪 Prediction on test set └── Generate submission.csv

---

## 📈 Evaluation Metric
**Pearson Correlation** used to evaluate prediction quality.

📌 Final Public Score: 0.519

---

## 📁 Files Included
- `Untitled0.ipynb` - Main notebook with code and explanations
- `submission.csv` - Output file with predictions for test set

---

## 💡 Future Enhancements

Mind Map - Improvements ├── Use advanced models (e.g. BERT, XGBoost) ├── Handle diverse accents ├── Use grammar-checking NLP tools └── Add audio-based features (e.g. fluency, pause detection)

---

## 👤 Author
**Crafted with care by [Avin Raj]** ✨

📬 For queries or collaborations, feel free to reach out!