---
license: cc-by-4.0
language:
- zh
tags:
- audio
- speech
---


### Dataset Description

This dataset consists of 55 audio recordings in Cantonese with the length between 1 seconds to 5 seconds of text extracted from a novel book on the theme of Sherlock Holmes. All recordings spoke in written language rather than spoken language in Cantonese and recorded using the application Praat. This dataset aim to provide audio and speech samples for the future training of the Text-to-Speech (TTS) models.





### Issues Encountered & Solution

I encountered problems during the preparation process. The first problem is the equipment related which alerts my attention when reviewing the sample of the testing samples. As mentioned in the instructions, recording should be in the range of 0.3 to 0.7 while my recording remained in 0.2. I had to buy a new microphone to record my recording as I have tried multiple ways to adjust my headphones but failed. Another problem is the background noise. With construction ongoing under my room and daily noise from my family, I had to record after 1 a.m. in order to have the suitable environment for recording.