README.md

metadata

title: SpeechEmotionNet
colorFrom: blue
colorTo: purple
sdk: streamlit
app_file: app.py
pinned: false
sdk_version: 1.51.0

SpeechEmotionNet – Speech Emotion Recognition App

SpeechEmotionNet is a real-time Speech Emotion Recognition (SER) system that predicts human emotions from audio using deep learning and MFCC-based feature extraction.

This app is built using:

• TensorFlow / Keras – Model training
• Librosa – Audio processing
• Scikit-Learn – Scaling + Label Encoding
• Streamlit – Web Interface
• HuggingFace Hub – Model hosting

---

Features

✓ Upload any .wav audio file
✓ Extract MFCC audio features
✓ Predict emotion using a trained neural network
✓ Shows probability distribution for all emotion classes
✓ Displays evaluation visuals (ROC curve, Confusion Matrix, CV Accuracy)
✓ Runs fully in-browser on CPU (no GPU required)

---

Emotion Classes

The model predicts one of the following emotions:

• Neutral
• Happy
• Sad
• Angry
• Fearful
• Surprised
• Calm
• Disgust