added last experiment results
Browse files
README.md
CHANGED
|
@@ -83,6 +83,18 @@ Novel architecture with separate P-GPT (phoneme) and S-GPT (syllable) decoders a
|
|
| 83 |
|
| 84 |
**Improvement**: 22% phoneme, 28% syllable error reduction
|
| 85 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
---
|
| 87 |
|
| 88 |
## Key Insights
|
|
|
|
| 83 |
|
| 84 |
**Improvement**: 22% phoneme, 28% syllable error reduction
|
| 85 |
|
| 86 |
+
#### Time Normalization Algorithm
|
| 87 |
+
| Version | Amplification Factor | Normal Spec CER (%) |
|
| 88 |
+
|-------------------------|----------------------|---------------------|
|
| 89 |
+
| Original + Noise Augment| - | 9.86 |
|
| 90 |
+
| Original | - | 11.51 |
|
| 91 |
+
| 1 | 2 | 9.93 |
|
| 92 |
+
| Random | Random | 9.97 |
|
| 93 |
+
| 8 (Updated Algo) | 2 | 12.77 |
|
| 94 |
+
| 9 (8 + Random Noise) | 2 | 15.33 |
|
| 95 |
+
| 10 | Random | 14.07 |
|
| 96 |
+
|
| 97 |
+
**Improvement**: 9.93% CER from original 11.51 for handicapped speaker
|
| 98 |
---
|
| 99 |
|
| 100 |
## Key Insights
|