KunquDB

• A large-scale, well-annotated audio-visual dataset
• Comprises 339 speakers and 128 hours of content
• Originated from the Kunqu Opera Art Canon, Kunqu yishu dadian
• Structured by dialogue lines, providing explicit annotations: character names, speaker names, gender, vocal manner classifications and preliminary text transcriptions

Kunqu yishu dadian

The Kunqu Opera Art Canon, encompasses the most significant literary, musical, and audiovisual materials spanning over 600 years, embodying the essence of Kunqu Opera art. Specifically, it contains over 22.3 million words of textual documentation, 396 sets of reprinted documents totaling over 70,000 pages, 127 hours of recorded audio, over 400 hours of video recordings, and over 6,000 images, all compiled into 149 volumes. For more details about this book, you can visit the official website of its publisher, or check out its introduction on douban.

Note

After purchasing the book, we negotiated with the publisher and secured their authorization for its utilization in Kunqu Opera research. The publisher explicitly stated that the book's digital resource can be employed solely for scholarly or research endeavors upon the approval of the publisher. It may not be illegally disseminated or used for commercial purposes.

Annotation Data Format

• Data structured by dialogue lines, metadata stored in a CSV table
• Contains all labels and annotations for each utterance
• Information for each utterance includes:
  • video ID indicates the corresponding video;
  • start and end timestamps specify the location of the utterance within the video;
  • character name denotes the character portrayed in the video, often associated with the role type, for the corresponding utterance;
  • performer name represents the individual performer portraying the character for the corresponding utterance;
  • vocal manner type categorizes the utterance into “stage speech” or “singing”, depending on how it’s vocalized;
  • preliminary content transcription corresponds to the transcription of the spoken content within the utterance.

Here’s a screenshot of the annotation:

Annotation Data Examples

Due to data restrictions, we choose not to publicly disclose the data from the KunquDB dataset as examples. Instead, we showcase similar annotations for online Kunqu Opera videos here.

Citation:

If our work is useful for your research, please consider citing:

Download

Researchers can gain access to the source video data by purchasing Kunqu yishu dadian. It is the user’s responsibility to get the approval from the publisher to conduct research for non-commercial purposes. We only provide our annotation dataset and processing scripts. To obtain the annotation dataset, please contact us via E-mail: huali.zhou@dukekunshan.edu.cn or ming.li369@dukekunshan.edu.cn, along with your affiliation and the consent from the publisher.

License

The dataset is licensed under the CC BY-NC-SA 4.0 license. This means that you can share and adapt the dataset for non-commercial purposes as long as you provide appropriate attribution and distribute your contributions under the same license. Detailed terms can be found on LICENSE.