Dataset: multiwoz_dst
================================

Short description
-----------------

Prepared MultiWOZ training dataset formatted for dialogue state tracking (DST) experiments with aligned audio and context fields.

Key metadata
------------

- Number of examples: 56,750
- Total size on disk: ~6.12 GB
 - Splits: train, validation
  - validation examples: 7,374
- Features:
  - `text` (string): original utterance text
  - `normalized_text` (string): normalized form of the utterance
  - `context` (string): preceding dialogue context
  - `normalized_context` (string): normalized context
  - `audio` (Audio, sampling_rate=16000): path/bytes for the audio
  - `audio_len` (int32): audio length
  - `wav_id` (string): wav file identifier
  - `turn_index` (int32): turn index within the dialogue
  - `agent_text` (string): agent/system utterance text (when applicable)
  - `domains` (string): domain labels
  - `slots` (string): slot annotation string

Credits
-------

Please credit the SpokenWoZ-Whisper preprocessing where applicable: https://huggingface.co/datasets/pirxus/spokenwoz-whisper

License & Citation
-------------------

- License: CC0 1.0 Universal (CC0-1.0)
- Paper reference / citation: https://arxiv.org/pdf/2511.22503

Notes
-----

- Add license/citation before publishing.
- Load locally with `datasets.load_from_disk` (see file paths in the dataset directory).