Mandarin tts dataset by ShenWanhong

# Mandarin TTS Dataset

## Dataset Information
- **Language**: Mandarin Chinese
- **Speaker**: Shen Wanhong
- **Student ID**: 11537606
- **Number of Sentences**: 50 sentences
- **Total Duration**: 10minutes
- **Recording Device**: Huawei Freebud5i earphone
- **Recording Environment**: Quiet bedroom
- **Audio Format**: WAV
- **Sample Rate**: 44.1 kHz
- **Channel**: Mono
- **Dataset Purpose**: Text-to-Speech (TTS) model training

## Dataset Preparation Reflection

During the preparation of this TTS dataset, I encountered two main technical challenges:

1. **Audio Segmentation Alignment Issue**: When using Audacity's automatic silence detection function, the system sometimes misidentifies breathing sounds as sentence boundaries. To address this, I adjusted the silence detection parameters, increasing the "minimum silence duration" from the default value to 0.5 seconds, and manually checked all sentence boundaries to ensure the accuracy of each audio segment.

2. **Background Noise Processing**: There was consistent, low-level background noise in the recording environment. I used Audacity's noise reduction function: first, I selected a segment of pure background noise as a sample, then applied the noise reduction effect with a strength setting of 12 dB. This method effectively reduced background noise while preserving speech clarity and naturalness.

Through this hands-on experience, I gained a deeper understanding of the importance of high-quality speech data for TTS model training and the impact of each preprocessing detail on the final data quality.

README.md

@@ -0,0 +1,15 @@
+---
+license: cc-by-4.0
+task_categories:
+- text-generation
+language:
+- zh
+tags:
+- speech
+- mandarin
+- audio
+- tts
+pretty_name: Mandarin tts dataset by ShenWanhong
+size_categories:
+- 1K<n<10K
+---