🎙️ Grammar Scoring Engine from Voice Samples

An end-to-end AI system that evaluates spoken English grammar by converting audio to text, detecting grammatical errors using NLP, generating a normalized score, and mapping it to CEFR proficiency levels.

---

🔗 Live Demo

👉 Try it live →

Upload any English .wav audio file and get instant grammar feedback!

---

🎯 Problem Statement

Assessing spoken grammar is critical for:

• Language learning platforms (Duolingo, IELTS prep)
• Hiring tools (spoken English screening)
• Educational assessments (CEFR-based evaluation)

This project builds a lightweight, cloud-deployable grammar evaluation pipeline that works on real-world audio input — no Java, no heavy dependencies.

---

🧠 System Architecture

Audio Input (.wav)
        │
        ▼
Speech-to-Text (Google Speech API)
        │
        ▼
Grammar Error Detection (LanguageTool NLP API)
        │
        ▼
Score Normalization (0–100)
        │
        ▼
CEFR Level Mapping (A1 → C2)
        │
        ▼
Streamlit Web Interface

---

⚙️ Tech Stack

Component	Technology
Language	Python 3.10+
Speech-to-Text	SpeechRecognition (Google Speech API)
Grammar Analysis	LanguageTool Public API (cloud-safe, Java-free)
UI & Deployment	Streamlit + Streamlit Cloud
Audio Samples	WAV format input

---

📊 Grammar Scoring Methodology

1. Spoken audio is transcribed into text using Google Speech API
2. Grammar errors are detected using rule-based NLP (LanguageTool)
3. Error density is computed against total word count
4. A normalized score (0–100) is generated
5. Score is mapped to CEFR proficiency level

🎯 CEFR Score Mapping

Score Range	CEFR Level	Description
90 – 100	C1 / C2	Advanced / Mastery
75 – 89	B2	Upper Intermediate
60 – 74	B1	Intermediate
40 – 59	A2	Elementary
< 40	A1	Beginner

---

✨ Features

• 🎤 Upload any English .wav audio file
• 📝 Real-time speech-to-text transcription
• 🔍 Automatic grammar error detection
• 📊 Normalized grammar score (0–100)
• 🏆 CEFR proficiency level estimation
• ✅ Highlighted error details
• ☁️ Fully cloud-deployed — works in browser, no setup needed

---

🚀 Run Locally

Step 1 — Clone the repo

git clone https://github.com/M1325-source/grammar-scoring-engine.git
cd grammar-scoring-engine

Step 2 — Install dependencies

pip install -r requirements.txt

Step 3 — Run the app

streamlit run app.py

Open http://localhost:8501 in your browser.

---

📁 Project Structure

grammar-scoring-engine/
│
├── app.py                  ← Streamlit web app (main entry)
├── train.py                ← Model training / scoring logic
├── requirements.txt        ← Python dependencies
├── audio_samples/          ← Sample .wav files for testing
└── README.md

---

🔍 Key Design Decisions

• Avoided Java-based LanguageTool — used REST API instead for Streamlit Cloud compatibility
• Dataset-agnostic pipeline — works on any real-world English audio
• Modular architecture — each stage (STT, NLP, scoring) is independent
• Cloud-first approach — no local server or runtime dependencies needed

---

🔮 Future Enhancements

• ML-based grammar scoring model (regression on CEFR labels)
• Pronunciation & fluency analysis
• Multi-language support
• Confidence & coherence metrics
• Real-time microphone recording
• Batch audio processing

---

👩‍💻 Author

Manisha Priya AI/ML Engineer | NLP Enthusiast

• 🐙 GitHub: @M1325-source
• 📧 Email: manishapriya1325@gmail.com

---

Built as part of an AI research assessment — demonstrates real-world NLP pipeline design, speech processing, and cloud deployment best practices.

⭐ If you found this useful, please star the repo!