OfficeDial Dataset

# OfficeDial Dataset

## EXPLANATION OF DATA FILES

We are releasing this dataset as a json file containing dialogues between a user and an IVA in different noise levels for different scenarios. The format of the dataset is adapted from [Taskmaster](https://github.com/google-research-datasets/Taskmaster) dataset.

The dataset is a dictionary of filenames and an array of conversations.

Each conversation contains the following attributes:

- conversation_id: a unique id

- scenario: scenario of this conversation, could be S1_A, S1_B, S2_A, S2_B, S3_A, S3_B

- noise: noise level played of during this conversation, values are SILENCE, NON_VERBAL, VERBAL

- utterances: an array of utterances

Each utterance contains the following fields:

- index: index representing the order of this conversation, starts at 0

- speaker: speaker of this specific line, values are USER, ASSISTANT

- text: The transcription of the spoken words

## License

Creative Commons Attribution License (cc-by).