Overview:

The Voice Emotion Detection project is an AI-based system that identifies human emotions (such as happiness, anger, sadness, fear, surprise, or neutrality) from voice recordings or real-time audio input.
Using speech signal processing, feature extraction (MFCC, Chroma, Spectral Contrast), and Machine Learning/Deep Learning models, the system analyzes tone, pitch, intensity, and rhythm of speech to determine the speaker’s emotional state.

This technology has wide applications in call centers, healthcare, human-computer interaction, sentiment analysis, and virtual assistants — making machines capable of understanding human emotions through voice.

Objectives:

• To develop an AI system that can detect and classify emotions from speech.

• To extract and analyze audio features for emotion recognition.

• To build and train a Machine Learning or Deep Learning model for classifying emotions.

• To demonstrate real-time voice emotion detection with an intuitive interface.

Key Features:

1. Emotion Classification: Detects multiple emotions such as happy, sad, angry, fear, calm, or neutral.

2. Audio Input Support: Accepts recorded voice clips or real-time microphone input.

3. Feature Extraction: Uses Mel Frequency Cepstral Coefficients (MFCC), Chroma, and Spectral features.

4. AI-Powered Model: Employs CNNs, RNNs, or LSTM networks for emotion classification.

5. Graphical Output: Displays emotion prediction with probability percentages.

6. Dataset Training: Trained on emotion datasets like RAVDESS, TESS, or SAVEE.

7. Interactive Interface: Simple web interface to record and analyze user speech.

8. Model Visualization: Shows accuracy and loss graphs during training.

9. Real-Time Processing: Instant prediction after user speaks or uploads audio.

10. Offline Capability: Can work without internet after model training.

Tech Stack:

• Frontend: HTML, CSS, Bootstrap, JavaScript (with Web Audio API for recording)

• Backend: Python (Flask / Django) / Node.js

• Machine Learning / Deep Learning:

  • Libraries: TensorFlow, Keras, Librosa, scikit-learn, NumPy, Pandas, Matplotlib

  • Techniques: Audio Feature Extraction, Classification

  • Models: CNN, RNN, LSTM, or Hybrid Deep Learning models

• Dataset:

  • RAVDESS (Ryerson Audio-Visual Database of Emotional Speech and Song)

  • TESS (Toronto Emotional Speech Set)

  • SAVEE (Surrey Audio-Visual Expressed Emotion Dataset)

• Database (optional): MySQL / Firebase (for storing results or user data)

Workflow:

1. Data Collection:

   • Collect pre-labeled emotional voice samples from datasets (e.g., RAVDESS).

2. Preprocessing:

   • Convert audio to mono and fixed sample rate.

   • Extract features like MFCC, Chroma, Spectral Centroid, and Zero-Crossing Rate.

3. Model Training:

   • Train ML/DL models using extracted features and labeled emotions.

   • Use algorithms like CNN, RNN, or SVM for classification.

4. Prediction Phase:

   • Record or upload a new audio sample.

   • Extract features and classify the emotion using the trained model.

5. Output Visualization:

   • Display detected emotion (e.g., “Happy ”) with confidence score.