EMOTION DETECTION FROM VOICE
USING ML
PRESENTED BY
RAAVI NEHA (212T1A04A3)
V.B.CHINNI (212T1A04C7)
N.SRI HARSHINI (212T1A0486)
E.SREE LAKSHMI (222T5A0402)
UNDER THE GUIDANCE OF
R.MERLIN (PH.D)
ASSOCIATE PROFESSOR
DEPT OF ECE
CONTENTS
ABSTRACT
INTRODUCTION
EXISTING METHOD
DRAWBACKS
PROPOSED METHOD
METHODOLOGY
CONCLUSION
APPLICATIONS
ABSTRACT
Emotion recognisation from voice is an emerging
field in artificial intelligence and human-
computer interaction . Our project aims to
develop a machine learning-based model to
detect emotions from human speech.
In our project, we aim to detect Emotion detection from Voice using
Machine Learning (ML).
INTRODUCTION
Our project creates an AI system to recognize emotions from speech .
people show emotions from speech . people show emotions through their
voice and AI can help detect them automatically.
Instead of humans analyzing speech, this system uses machine learning
to do it faster. It takes important voice features (MFCCs) and trains a CNN
model to detect emotions like happiness, sadness, anger, and fear.
This model learns from different voices to improve accuracy. This can
help in customer service, virtual assistants, and mental health support.
Future goals : Make it faster, more accurate, and work with all voices.
EXISTING METHOD
• It can be classify in to three main categories :
1. Rule-based
approaches
2. Statistical models
3. Machine learning
techniques
RULE-BASED & STATISTICAL APPROACHES
• Early systems relied on predefined acoustic features like pitch, energy, and speech rate.
• Common statistical models:
HMMs (Hidden Markov Models): Good for sequential data but struggled with speech variability.
GMMs (Gaussian Mixture Models): Needed large labeled datasets for good performance.
SVMs (Support Vector Machines): Accurate but required extensive feature engineering.
Limitations :
• Poor adaptability to different speakers, background noise, and accents.
• Heavy reliance on manually selected features.
MACHINE LEARNING-BASED APPROACHES
Introduced automated feature extraction and improved classification.
Key techniques:
KNN: Simple but computationally expensive for large datasets.
ANNs: Learned features well but needed large labeled data.
RNNs & LSTMs: Captured speech patterns but were hard to train.
Limitations:
• Required extensive preprocessing and domain expertise.
• Struggled with generalizing to new datasets.
EXISTING METHOD DRAWBACKS
Dependency on Handcrafted Features
Poor Generalization across speakers
Sensitivity to Background Noise
Computational complexity & Real-time constraints
Limited availability of High-Quality labeled data
Lack of context Awareness
Difficulty in handling Mixed Emotions
Ethical & privacy concerns
PROPOSED METHOD
• To overcome the limitations of existing emotion detection systems, we propose a deep
learning-based approach that leverages Convolutional Neural Networks (CNNs) to
classify emotions from speech signals
• Key Components of the System
Deep Learning for Feature Extraction
CNN-Based Emotion Classification
Speaker Independence & Generalization
Real-Time Emotion Detection
METHODOLOGY
Data Collection
Preprocessing
Feature
Extraction
Model Evaluation
Emotion
Prediction
Display Results
Deployment
Applications
CONCLUSION
• This project detects emotions from voice using machine learning. It
listens to speech, extracts important features, and uses a deep
learning model (CNN) to recognize emotions. This method is more
accurate than older techniques.
• Key Takeaways are : accuracy & performance gains , Generalization
to new Data , Real-time prediction capability and Overcoming
Traditional limitations.
APPLICATIONS
Customer Service & Call Centers .
Healthcare & Mental Health Monitoring.
Education & E-learning.
Entertainment & Gaming.
Human-Computer Interaction (HCI) & Virtual Assistants.
Security & Forensic Applications.
Marketing & Consumer Behaviour Analysis.
