Update README.md
Browse files
README.md
CHANGED
|
@@ -25,6 +25,14 @@ For labeled data, please refer to our [HuggingFace](https://huggingface.co/datas
|
|
| 25 |
|
| 26 |
For full dataset (labeled data + unlabeled data), please refer to [Google Drive](https://drive.google.com/drive/folders/1hsoB_xjWh66glKg3tQaSLm4S1SVPyANP?usp=sharing)
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
## Contact:
|
| 29 |
|
| 30 |
If any links are broken, please contact me for fixing!
|
|
|
|
| 25 |
|
| 26 |
For full dataset (labeled data + unlabeled data), please refer to [Google Drive](https://drive.google.com/drive/folders/1hsoB_xjWh66glKg3tQaSLm4S1SVPyANP?usp=sharing)
|
| 27 |
|
| 28 |
+
## Limitations:
|
| 29 |
+
|
| 30 |
+
Since this dataset is human-labeled, 1-2 ending/starting words present in the recording might not be present in the transcript.
|
| 31 |
+
That's the nature of human-labeled dataset, in which humans can't distinguish words that are faster than 1 second.
|
| 32 |
+
In contrast, forced alignment could solve this problem because machines can "listen" words in 10ms-20ms.
|
| 33 |
+
However, forced alignment only learns what it is taught by humans.
|
| 34 |
+
Therefore, no transcript is perfect. We will conduct human-machine collaboration to get "more perfect" transcript in the next paper.
|
| 35 |
+
|
| 36 |
## Contact:
|
| 37 |
|
| 38 |
If any links are broken, please contact me for fixing!
|