Update README.md
Browse files
README.md
CHANGED
|
@@ -69,11 +69,8 @@ The model was fine-tuned using the **ScreenTalk-XS dataset**, a collection of tr
|
|
| 69 |
---
|
| 70 |
|
| 71 |
## 📊 Evaluation
|
| 72 |
-
### 📌 Testing Data
|
| 73 |
-
- **Dataset**: ScreenTalk-XS
|
| 74 |
-
- **Metrics**: Word Error Rate (WER)
|
| 75 |
|
| 76 |
-
### 📌 Results
|
| 77 |
| **Epoch** | **Training Loss** | **Validation Loss** | **WER (%)** |
|
| 78 |
|-----------|-----------------|-----------------|-------------|
|
| 79 |
| **1** | 0.502400 | 0.333292 | 20.870653 |
|
|
@@ -85,6 +82,17 @@ The model was fine-tuned using the **ScreenTalk-XS dataset**, a collection of tr
|
|
| 85 |
- **Best Model:** `Epoch 4`, achieving **WER = 20.50%**
|
| 86 |
- **Model performance degrades after epoch 4**, suggesting overfitting.
|
| 87 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 88 |
---
|
| 89 |
|
| 90 |
## 🖥️ Technical Specifications
|
|
|
|
| 69 |
---
|
| 70 |
|
| 71 |
## 📊 Evaluation
|
|
|
|
|
|
|
|
|
|
| 72 |
|
| 73 |
+
### 📌 Training Results
|
| 74 |
| **Epoch** | **Training Loss** | **Validation Loss** | **WER (%)** |
|
| 75 |
|-----------|-----------------|-----------------|-------------|
|
| 76 |
| **1** | 0.502400 | 0.333292 | 20.870653 |
|
|
|
|
| 82 |
- **Best Model:** `Epoch 4`, achieving **WER = 20.50%**
|
| 83 |
- **Model performance degrades after epoch 4**, suggesting overfitting.
|
| 84 |
|
| 85 |
+
### 📌 Test Results
|
| 86 |
+
| **Model** | **WER (%)** |
|
| 87 |
+
|-----------|-------------|
|
| 88 |
+
| **Whisper-small (baseline)** | **30.00%** |
|
| 89 |
+
| **ScreenTalk-xs (fine-tuned)** | **27.00%** ✅ |
|
| 90 |
+
|
| 91 |
+
### **🔍 Key Observations**
|
| 92 |
+
- **Fine-tuning reduced WER from 30.00% → 27.00%** 🎯
|
| 93 |
+
- **Achieved a 10% relative improvement in ASR accuracy.**
|
| 94 |
+
- **Tested on the ScreenTalk-XS dataset.**
|
| 95 |
+
|
| 96 |
---
|
| 97 |
|
| 98 |
## 🖥️ Technical Specifications
|