eduhk-compling/11618204_KAUSARHUMERA

Dataset Description

This dataset consists of a single audio recording intended for Text-to-Speech (TTS) training. The recording is approximately 3 minutes long and features clear spoken content recorded in a quiet environment, although some background noise was present, which I attempted to reduce.

Audio File Details

File Name: s001.wav Duration: 3 minutes Content: The audio includes a reading of various sentences, all of which are fully transcribed in the accompanying metadata.csv file. Process and Challenges

During the preparation of this dataset, I encountered several challenges:

Segmentation and Labeling:

I initially aimed to segment the audio recording into smaller clips based on silence, using the Silence Finder tool in Audacity. However, I found it difficult to understand how to use the labeling function effectively. After multiple attempts, I realized I could not achieve the desired segmentation due to my unfamiliarity with Audacity's features. This led me to decide to use the entire recording instead of dividing it into smaller segments. Background Noise:

While recording, some ambient noises interfered with the clarity of the audio. Even after recording in a quieter space, I found that some noise was still present. I used Audacity to cut out sections with noticeable noise, but this did not eliminate all distractions. In light of these challenges, I opted to provide a single complete recording to maintain the integrity of the content instead of risking loss in quality from poor segmentation.

How to Use This Dataset

Download the audio file and the metadata. Utilize the audio and corresponding transcripts for TTS model training or other research purposes.