README.md

metadata

license: other
language:
  - en
tags:
  - audio
  - speech
  - speaker
  - evaluation
configs:
  - config_name: improvised
    data_files:
      - split: train
        path: improvised_train*.parquet
      - split: dev
        path: improvised_dev*.parquet
      - split: test
        path: improvised_test*.parquet
  - config_name: naturalistic
    data_files:
      - split: train
        path: naturalistic_train*.parquet
      - split: dev
        path: naturalistic_dev*.parquet
      - split: test
        path: naturalistic_test*.parquet
task_categories:
  - audio-classification
pretty_name: seamless-interaction

seamless-interaction

Folder structure

root/
  |- audio_000/
  |- audio_001/
  |- data_000.parquet
  |- data_001.parquet

Column

{
  "conversation_id": "{conversation_id}",
  "utterances": [
    {
      "spk": 0, // int
      "words": [
        {
          "word": "str", // str
          "start_time": 0.0, // float
          "end_time": 0.08 // flat
        }
      ],
      "label": "improvised/naturalistic",
      "split": "train/dev/test",
      "speaker_id": {"0": "xxxx", "1": "xxxx"}
    }
  ],
  "audio_path": "audio_{group_idx}/{conversation_id}.wav"
}