Update README.md
Browse files
README.md
CHANGED
|
@@ -66,21 +66,11 @@ These visualizations help confirm that:
|
|
| 66 |
- The signal is strong enough to potentially help the model understand pitch-sensitive aspects of speech
|
| 67 |
|
| 68 |
----
|
| 69 |
-
#### Domain-Specific ASR
|
| 70 |
|
| 71 |
#### freqs = (theta / 220.0) * 700 * (torch.pow(10, torch.linspace(0, 2595 * torch.log10(torch.tensor(1 + 8000/700)), dim // 2, device=device, dtype=dtype) / 2595) - 1) / 1000
|
| 72 |
|
| 73 |
-
#### Static frequency's are perfectly fine for text models but not for NLP
|
| 74 |
-
-----
|
| 75 |
-
|
| 76 |
-
1000 steps no f0:
|
| 77 |
-
|
| 78 |
-
<img width="470" alt="123" src="https://github.com/user-attachments/assets/1b3ca1e8-0b7d-47dd-802b-5eda9537ae13" />
|
| 79 |
-
|
| 80 |
-
1000 steps with f0 / theta substitutions:
|
| 81 |
-
|
| 82 |
-
<img width="470" alt="65356" src="https://github.com/user-attachments/assets/84624fc4-5def-4e9f-9cdd-c350b80ec348" />
|
| 83 |
-
|
| 84 |
|
| 85 |
----
|
| 86 |
|
|
|
|
| 66 |
- The signal is strong enough to potentially help the model understand pitch-sensitive aspects of speech
|
| 67 |
|
| 68 |
----
|
| 69 |
+
#### Domain-Specific ASR/NLP.
|
| 70 |
|
| 71 |
#### freqs = (theta / 220.0) * 700 * (torch.pow(10, torch.linspace(0, 2595 * torch.log10(torch.tensor(1 + 8000/700)), dim // 2, device=device, dtype=dtype) / 2595) - 1) / 1000
|
| 72 |
|
| 73 |
+
#### Static frequency's are perfectly fine for text models but not for NLP
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
|
| 75 |
----
|
| 76 |
|