Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -8,11 +8,11 @@ tags:
|
|
| 8 |
---
|
| 9 |
|
| 10 |
### Dataset Description
|
| 11 |
-
|
| 12 |
This dataset contains high-quality voice recordings of 40 sentences selected from the famous Chinese prose "Moonlight over the Lotus Pond" (荷塘月色) by Zhu Ziqing. The total duration of the audio is over 3 minutes, designed specifically for Text-to-Speech (TTS) model fine-tuning or linguistic analysis.
|
| 13 |
|
| 14 |
### Issues Encountered & Solutions
|
| 15 |
-
|
| 16 |
During the dataset preparation, I encountered two main issues:
|
|
|
|
| 17 |
First, there was significant **background noise and breath sounds** in the initial recordings, which affected the clarity of the prose's delicate tone. I resolved this by applying "Noise Reduction" in Audacity and manually fading out the breathing gaps.
|
|
|
|
| 18 |
Second, I faced **segmentation alignment errors** due to the long, rhythmic sentences characteristic of Zhu Ziqing’s style. Some transcriptions didn't perfectly match the natural pauses in my speech. I resolved this by re-listening to each clip and carefully adjusting the text in the `metadata.csv` to ensure every word and pause aligned precisely with the audio.
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
### Dataset Description
|
|
|
|
| 11 |
This dataset contains high-quality voice recordings of 40 sentences selected from the famous Chinese prose "Moonlight over the Lotus Pond" (荷塘月色) by Zhu Ziqing. The total duration of the audio is over 3 minutes, designed specifically for Text-to-Speech (TTS) model fine-tuning or linguistic analysis.
|
| 12 |
|
| 13 |
### Issues Encountered & Solutions
|
|
|
|
| 14 |
During the dataset preparation, I encountered two main issues:
|
| 15 |
+
|
| 16 |
First, there was significant **background noise and breath sounds** in the initial recordings, which affected the clarity of the prose's delicate tone. I resolved this by applying "Noise Reduction" in Audacity and manually fading out the breathing gaps.
|
| 17 |
+
|
| 18 |
Second, I faced **segmentation alignment errors** due to the long, rhythmic sentences characteristic of Zhu Ziqing’s style. Some transcriptions didn't perfectly match the natural pauses in my speech. I resolved this by re-listening to each clip and carefully adjusting the text in the `metadata.csv` to ensure every word and pause aligned precisely with the audio.
|