Update README.md
Browse files
README.md
CHANGED
|
@@ -21,6 +21,8 @@ Dataset used for finetuning is Common Voice 17.0, Russian part, that contains ov
|
|
| 21 |
After preprocessing of the original dataset (all splits were mixed and splited to a new train + test split by 0.95/0.05,
|
| 22 |
that is 225761/11883 rows respectively) the original Whisper v3 has WER 9.84 while the finetuned version shows 6.39 (so far).
|
| 23 |
|
|
|
|
|
|
|
| 24 |
## Usage
|
| 25 |
|
| 26 |
```python
|
|
|
|
| 21 |
After preprocessing of the original dataset (all splits were mixed and splited to a new train + test split by 0.95/0.05,
|
| 22 |
that is 225761/11883 rows respectively) the original Whisper v3 has WER 9.84 while the finetuned version shows 6.39 (so far).
|
| 23 |
|
| 24 |
+
Finetuning process took over 60 hours on dual Tesla A100 80Gb.
|
| 25 |
+
|
| 26 |
## Usage
|
| 27 |
|
| 28 |
```python
|