Datasets:
license: cc-by-4.0
language:
- zh
tags:
- speech
- audio
Dataset Description
This dataset includes 50 sentences spoken in colloquial Hong Kong Cantonese (HKC), covering interrogatives and statements. Sentences are sourced from online Cantonese teaching materials and classical commercial slogans. It includes approxiametly 198 seconds of audio recorded by a female native speaker of HKC. The sampling rate is 44.1 kHz with 16-bit resolution. Transcription in Jyutping were also provided.
Issues Encountered & Solution
I encountered two issues when preparing the dataset. First, since the speaker exhibits some ongoing sound changes (e.g. initial n-l variation) in her speech, I had to decide how to treat pronunciation variants. To maintain naturalness and preserve the speaker's unique pronunciation and speech style, I chose not to consciously modify my pronunciation during recording. Second, it was difficult to ensure that the 50 sentences covered every possible Cantonese sounds and combination while keep the content natrual and meaningful. Given the small size of the dataset, full phonetic coverage was not prioritized.