Change README.md

README.md

@@ -7,72 +7,49 @@ tags:
 - speech
 - meld
 - wav2vec2
-- pooling
----
-# Audio Emotion Recognition Model
-
-This repository contains an audio-based emotion recognition model trained on the MELD dataset.
-The model is designed as a strong unimodal baseline and is later used as a component in a
-multimodal emotion recognition system.
-
+- temporal-pooling
 ---
 
-## Model Overview
+# Audio Emotion Recognition Model
 
-- **Task:** Speech Emotion Recognition
-- **Dataset:** MELD
-- **Backbone:** Wav2Vec 2.0 (pretrained)
-- **Pooling Strategy:** Temporal mean pooling
-- **Classifier:** MLP with class-weighted loss
+An audio-based emotion recognition model trained on the **MELD** dataset.  
+It serves as a strong **unimodal audio baseline** and as the audio encoder in a multimodal emotion recognition system.
 
-The model processes raw audio signals and predicts one of the seven emotion classes defined
-in the MELD dataset.
+## Model Summary
 
----
+- **Task:** Speech Emotion Recognition  
+- **Dataset:** MELD  
+- **Backbone:** `facebook/wav2vec2-base`  
+- **Pooling:** Temporal pooling (mean + std over time)  
+- **Classifier:** MLP with class-weighted loss  
+- **Classes:** 7 emotion categories  
 
 ## Architecture
 
-1. **Feature Extraction**
-   - Pretrained `facebook/wav2vec2-base`
-   - Outputs frame-level hidden representations
+1. **Wav2Vec 2.0 Encoder**  
+   Extracts frame-level representations from raw audio.
 
-2. **Temporal Pooling**
-   - Mean pooling across the time dimension
-   - Converts variable-length sequences into fixed-size utterance embeddings
+2. **Temporal Pooling**  
+   Mean and standard deviation pooling over the time dimension to obtain a fixed-size utterance embedding.
 
-3. **Classification Head**
-   - Fully connected MLP
-   - Dropout regularization
-   - ReLU activation
-   - Softmax output layer
+3. **MLP Classifier**  
+   Fully connected layers with ReLU and dropout, followed by a softmax output layer.
 
----
+## Class Imbalance Handling
 
-## Handling Class Imbalance
-
-Due to significant class imbalance in the MELD dataset, **class weights** are applied in the
-cross-entropy loss function.  
-This improves performance on underrepresented emotion classes and leads to better macro-F1
-scores.
-
----
+Class imbalance in MELD is addressed using **class weights** in the cross-entropy loss, improving macro-level performance on underrepresented emotions.
 
 ## Training Details
 
-- Sampling rate: 16 kHz
-- Maximum utterance duration: 6 seconds
-- Optimizer: AdamW
-- Loss function: CrossEntropyLoss with class weights
-- Evaluation metrics:
-  - Accuracy
-  - Macro F1-score
-  - Weighted F1-score
-
----
+- Sampling rate: 16 kHz  
+- Max utterance length: 6 seconds  
+- Optimizer: Adam
+- Loss: CrossEntropyLoss (with class weights)  
+- Metrics: Accuracy, Macro F1, Weighted F1  
 
 ## Usage
 
-This model serves as:
-- A standalone audio emotion classifier
-- An audio encoder in a multimodal emotion recognition framework
+- Standalone audio emotion classifier  
+- Audio branch for early and late fusion in multimodal emotion recognition
+