Skip to main content

Dataset Specification

Data Distribution and Installation​

The data is distributed as two gzipped tar archives:

  • cadenza_clip1_data.train.v1.0.tar.gz [4.5GB]: labelled training data.
  • cadenza_clip1_data.valid.v1.0.tar.gz [576MB]: unlabelled validation data.
Demo data

If you need to have a quick view of the data structure, we also provide a demo package:

  • cadenza_clip1_data.demo.v1.0.tar.gz [4MB]

This package contains 5 samples from the training and 5 from the validation data.

- Installation Instructions

  1. Download the .tar.gz files.
  2. unpack each archive using the following commands:
tar -xvzf cadenza_clip1_data.train.v1.0.tar.gz  # For training data
tar -xvzf cadenza_clip1_data.valid.v1.0.tar.gz # For validation data

- Directory Structure after unpacking

cadenza_data/
├── metadata/
├── train/
│   ├── signals/      # Audio (1) to predict intelligibility
│   └── unprocessed/  # Audio (2) without hearing loss
├── valid/
│   ├── signals/
│   └── unprocessed/
└── Manifest/

Audio signals​

  • Stored as 16-bit stereo FLAC files at 44100 Hz.
  • Filenames:
    • train/signals/<HASH_NUMBER>.flac: audio (1) signal to predict intelligibility.
    • train/unprocessed/<HASH_NUMBER>_unproc.flac: the unprocessed (without hearing loss) audio (2) signal.
  • Notes:
    • Audios (1) and unprocessed audios (2) have matching <HASH_NUMBER>.
    • Slight misalignment and variations in the number of frames may occur between the signals audio and the unprocessed audio due to the hearing loss simulation.

Training Metadata​

The train metadata is saved in metadata/train_metadata.json. The metadata contains list of dictionaries, each representing the listener's response.

Fields:

  • signal: audio filename.
  • prompt: original target sentence (ground truth).
  • response: text transcription from Prolific listening tests ('#' indicates no words entered).
  • n_words: number of words in prompts after expanding contractions (see data construction).
  • words_correct: number of correctly identified words.
  • correctness: Intelligibility score i.e. rate of correctly identified words (the target variable).
  • hearing_loss: Indicating if the signal audio was: not processed (No Loss); has Mild simulated hearing loss; or Moderate simulated hearing loss
metadata/train_metadata.json
[
  {
    "signal": "b2486a8e181f84acd2db4d61",
    "prompt": "in my mind I can see is fine",
    "response": "#",
    "n_words": 8,
    "words_correct": 0,
    "correctness": 0.0,
    "hearing_loss": "Moderate"
  }
]

Validation Metadata​

The validation metadata is saved in metadata/valid_metadata.json. The metadata contains list of dictionaries, each representing the listener's response.

Fields:

  • signal: name of signal to predict intelligibility from.
  • prompt: original target sentence (ground truth).
  • n_words: number of words in prompts after expanding contractions.
  • hearing_loss: Indicating if the signal audio was: not processed (No Loss); has Mild simulated hearing loss; or Moderate simulated hearing loss
cadenza_data/metadata/valid_metadata.json
[
  {
    "signal": "c657dc0bd1f822f8ff7104a1",
    "prompt": "do you know i am truly alone",
    "n_words": 7,
    "hearing_loss": "Moderate"
  }
]