The Nursing Handover ASR Speech Dataset is a multi-speaker clinical speech corpus designed for research in Automatic Speech Recognition (ASR) and speech-driven clinical documentation. The dataset contains nursing handover statements spoken by speakers with different English accents, enabling research on accent variability and domain adaptation in medical speech recognition systems.

This dataset supports research in:

Medical and clinical ASR
Speech-to-text automation in healthcare
Accent robustness evaluation
Domain-specific speech model fine-tuning
Clinical communication analysis

Data Source and Preparation

The nursing handover text used in this dataset was derived from the Synthetic Nursing Handover Training and Development Dataset provided by CSIRO.

For the Australian English speaker, the transcripts available in the CSIRO portal were not fully aligned with the corresponding audio recordings. Therefore, the audio–text alignment was manually corrected and aligned by the authors to ensure accurate pairing for ASR training and evaluation.

The Indian English (S1 and S2) recordings were independently recorded by the authors, using the same nursing handover scripts to maintain consistency across speakers while enabling accent variability analysis.

Speakers

The dataset consists of recordings from three speakers:

Australian Speaker
- Aus_D1
- Aus_D2
Indian Speaker S1
- S1D1
- S1D2
Indian Speaker S2
- S2D1
- S2D2

Each speaker includes two datasets representing separate recording sessions.

Dataset Description

The speech recordings consist of structured nursing handover statements commonly used in hospital environments. The recordings are suitable for training and evaluating ASR systems in healthcare settings.

Indian Speaker S1

Contains nursing handover statements with relatively longer duration recordings.
Designed for model training and evaluation under extended speech conditions.

Indian Speaker S2

Contains relatively shorter and more concise handover recordings.
Suitable for evaluating model robustness across varying speech lengths.

Medical Keywords Data

Contains medical words for each corressponding ground truth.
Human-in-loop vocabulary creation.
Need for human-in-loop vocabulary creation because existing clinical LLMs cannot detect nursing-domain-specific vocabularies like jargons or abbreviations.

Dataset Statistics

Speaker	Dataset	No. of Audio Files	Minimum Duration (s)	Maximum Duration (s)	Total Duration (minutes)	Min/Max Words	Total Words
Australian English	D1	100	15.88	97.05	68.07	31 / 222	8428
Australian English	D2	100	11.15	61.02	44.64	30 / 157	6958
Indian English (S1)	D1	100	16.55	126.68	78.53	31 / 222	8428
Indian English (S1)	D2	100	16.00	84.63	65.81	30 / 157	6958
Indian English (S2)	D1	100	12.27	107.99	62.58	31 / 222	8428
Indian English (S2)	D2	100	13.23	74.47	52.70	30 / 157	6958

Medical Keyword - Dataset Statistics

Dataset	Total Medical Words	Unique Words
D1	846	319
D2	819	296

Folder Structure

Nursing-Handover-ASR-Speech-Dataset/
│
├── Australian Speaker
│   ├── Aus_D1/
│   ├── Aus_D2/
│  
├── Indian Speaker S1
│   ├── S1D1/
│   ├── S1D2/
│   
├── Indian Speaker S2
│   ├── S2D1/
│   ├── S2D2/
│
├── Text/
│
├── Medical Keywords Data
│
└── README.md

The Speakers folder contains audio recordings grouped by speaker and dataset.
The Text folder contains corresponding transcription files.
Readme.md contains the details of the dataset.

Audio Specifications

Format: WAV
Sampling Rate: 16 kHz
Channels: Mono
Language: English
Domain: Clinical Nursing Handover

Intended Use

This dataset is suitable for:

Training ASR models such as Whisper, Wav2Vec2, HuBERT, and XLSR
Evaluating speech recognition performance in clinical environments
Accent generalization studies
Speech-driven clinical documentation research

Citation

If you use this dataset in your research, please cite:

@dataset{nursing_handover_asr_speech_dataset,
  title = {Nursing Handover ASR Speech Dataset},
  author = {Sasikala D, Siva Sathya S, Niranjan Kumar D, Vignesh S},
  year = {2026},
  url = {https://huggingface.co/datasets/vignesh242005/Nursing-Handover-ASR-Speech-Dataset},
  huggingface dataset id = {vignesh242005/Nursing-Handover-ASR-Speech-Dataset}
}

CSIRO Dataset Reference

@dataset{CSIRO,
  author = {Angel, Maricel and Suominen, Hanna and Zhou, Liyuan and Hanlen, Leif},
  title = {Synthetic nursing handover training and development data set - audio files},
  year = {2014},
  publisher = {CSIRO},
  version = {v1},
  doi = {10.4225/08/58d0977ab4888}
}

License

This dataset is provided for research and academic use only.
Commercial use requires permission from the authors.

Acknowledgement

This dataset was developed to support research in improving speech recognition systems for healthcare communication and reducing documentation workload in clinical environments.

Downloads last month: -

You need to agree to share your contact information to access this dataset

Nursing Handover ASR Speech Dataset

Overview