🐾 Animal Translator — NLP Project

How It Works

Animal Sound (.wav)
        ↓
[Audio Classifier]  — trained on ESC-50 (Kaggle)
        ↓ predicts emotion label e.g. "anger", "sadness", "joy"
[NLP Emotion→Sentence Model] — trained on Emotions NLP dataset (Kaggle)
        ↓
"Back off! This is my space and I mean it!"

Both stages are trained purely on Kaggle datasets. No hand-crafted rules. No APIs.

---

📥 Datasets to Download (BOTH required)

Dataset 1 — Audio (for the classifier)

ESC-50: Environmental Sound Classification 🔗 https://www.kaggle.com/datasets/mmoreaux/environmental-sound-classification-50

Download and place files:

data/
├── audio/        ← all .ogg files from ESC-50
└── esc50.csv     ← metadata CSV from ESC-50

Dataset 2 — NLP Text (for the translator)

Emotions Dataset for NLP by praveengovi 🔗 https://www.kaggle.com/datasets/praveengovi/emotions-dataset-for-nlp

Download and place files:

data/
├── train.txt     ← NLP training data  (format: "text;emotion")
├── val.txt       ← NLP validation data
└── test.txt      ← NLP test data

---

Setup

pip install -r requirements.txt

Run in Order

# Step 1: Train the audio classifier on ESC-50
python step1_train_audio_classifier.py

# Step 2: Train the NLP translator on Emotions dataset
python step2_train_nlp_translator.py

# Step 3: Translate an animal sound end-to-end
python step3_translate.py --audio path/to/sound.wav

# Step 4: Evaluate with BLEU + accuracy
python step4_evaluate.py

---

The NLP Connection

The Emotions for NLP dataset contains 20,000 sentences like:

"i feel absolutely helpless right now"     → sadness
"im so excited i could burst"              → joy
"i want to destroy everything around me"   → anger

Animals experience the same emotions. So:

• Dog aggressive bark → anger → model generates an angry sentence in animal voice
• Cat purring → joy → model generates a joyful sentence
• Dog whining alone → sadness → model generates a sad/lonely sentence

The NLP model learns from 20,000 real human sentences what each emotion sounds like, then generates translations in an animal's voice.