Hokchia Audio Dataset

Hokchia, or the Fuqing dialect, is a branch of Eastern Min Chinese spoken mainly in the Fuqing City of Fujian province, China. Unlike Hokkien, which is more widely recognized and spoken in various parts of Southeast Asia, Hokchia maintains its unique linguistic characteristics and is primarily used within the Fuqing community and its diaspora. This dialect is known for its distinct pronunciation, vocabulary, and grammatical structures compared to other Min Chinese varieties. The Hokchia audio dataset is designed to foster speech recognition technologies that cater to this unique dialect, aiming to enhance digital inclusivity for Hokchia speakers worldwide.

Dataset Description

The Hokchia Audio Dataset is a collection of audio recordings in the Hokchia language, accompanied by transcriptions. It is intended for use in speech recognition models, particularly to fine-tune models like Whisper for the Hokchia language. The dataset includes a wide range of spoken content, making it suitable for various applications requiring speech-to-text capabilities in Hokchia.

Content

Each audio file in the dataset is named following the pattern Hokchia_X.wav, where X is a numerical identifier. Accompanying each audio file is a JSON line in the whisper_finetune_input.jsonl file, providing the text transcription of the audio content. The dataset structure is as follows:

• README.md: This file.
• dataset/: Directory containing audio files split into subdirectories by language.
  • Hokchia/: Subdirectory containing Hokchia audio files.
• whisper_finetune_input.jsonl: JSON Lines file containing mappings of audio file paths to their text transcriptions.

Dataset Structure

Each line in the whisper_finetune_input.jsonl file represents a data point in the following format:

{"audio_filepath": "./dataset/Hokchia/Hokchia_1.wav", "text": "text transcription here"}

audio_filepath: Relative path to the audio file. text: Transcription of the audio in Hokchia. Use Cases

This dataset is particularly suited for:

Training and fine-tuning speech recognition models on the Hokchia language. Linguistic studies focusing on the Hokchia dialect. Developing voice-activated applications that require understanding of Hokchia. How to Use

You can load this dataset using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("AnnoFichel/hokchia_audio_dataset")

Acknowledgements

This dataset was collected and prepared by Jack Huang. We acknowledge the contributions of the speakers who participated in the recording sessions and the individuals who provided transcriptions.