Update README.md
Browse filesAdd third-party property rights info
README.md
CHANGED
|
@@ -52,6 +52,7 @@ widget:
|
|
| 52 |
We introduce FinVoc2Vec, a vocal tone classifier designed for real-world corporate disclosures.
|
| 53 |
In the first stage, we apply a self-supervised pre-training procedure that allows the base model to adapt to the acoustic characteristics of disclosure environments using a sample of 500,000 unlabeled sentences of conference call speech. In the second stage, we apply a supervised fine-tuning procedure that enables the model to learn representations of human-labeled vocal tone. We construct a speech corpus containing
|
| 54 |
5,000 audio recordings of linguistically neutral sentences from conference calls and manually label each sentence with perceived vocal tone — positive, negative, or neutral.
|
|
|
|
| 55 |
|
| 56 |
## Example using a demo dataset
|
| 57 |
```python
|
|
|
|
| 52 |
We introduce FinVoc2Vec, a vocal tone classifier designed for real-world corporate disclosures.
|
| 53 |
In the first stage, we apply a self-supervised pre-training procedure that allows the base model to adapt to the acoustic characteristics of disclosure environments using a sample of 500,000 unlabeled sentences of conference call speech. In the second stage, we apply a supervised fine-tuning procedure that enables the model to learn representations of human-labeled vocal tone. We construct a speech corpus containing
|
| 54 |
5,000 audio recordings of linguistically neutral sentences from conference calls and manually label each sentence with perceived vocal tone — positive, negative, or neutral.
|
| 55 |
+
We cannot make the labeled audio data publicly available due to third-party property rights. However, interested researchers with an active license may contact us directly to explore potential access options.
|
| 56 |
|
| 57 |
## Example using a demo dataset
|
| 58 |
```python
|