Compare speed against NeMo
Browse files
README.md
CHANGED
|
@@ -53,11 +53,11 @@ scriptrs runs NVIDIA's weights unchanged and gets the same word error rate as NV
|
|
| 53 |
|
| 54 |
LibriSpeech test-clean, English (5.4 h), one utterance at a time:
|
| 55 |
|
| 56 |
-
| | Speed (x real time) | WER |
|
| 57 |
-
|---|---|---|
|
| 58 |
-
| **scriptrs CUDA** | **1534x** | **1.93 %** |
|
| 59 |
-
| NVIDIA NeMo (PyTorch) | 477x | 1.93 % |
|
| 60 |
-
| parakeet-rs 0.4.3 (ONNX Runtime) | 253x | 1.93 % |
|
| 61 |
|
| 62 |
### Multilingual
|
| 63 |
|
|
|
|
| 53 |
|
| 54 |
LibriSpeech test-clean, English (5.4 h), one utterance at a time:
|
| 55 |
|
| 56 |
+
| | Speed (x real time) | vs NeMo | WER |
|
| 57 |
+
|---|---|---|---|
|
| 58 |
+
| **scriptrs CUDA** | **1534x** | **3.2x** | **1.93 %** |
|
| 59 |
+
| NVIDIA NeMo (PyTorch) | 477x | 1x | 1.93 % |
|
| 60 |
+
| parakeet-rs 0.4.3 (ONNX Runtime) | 253x | 0.53x | 1.93 % |
|
| 61 |
|
| 62 |
### Multilingual
|
| 63 |
|