SAA-Lab/SLPHelmManualLabels

SLP Helm Manual Labels

Dataset Summary

SLP Helm Manual Labels is a dataset containing speech disorder classifications with audio files and manual transcriptions. The dataset is designed for speech-language pathology research and includes both healthy control samples and samples from individuals with various speech disorders.

Supported Tasks and Leaderboards

• Speech Disorder Classification: Binary classification between healthy speech and speech disorders
• Speech Disorder Type Classification: Multi-class classification of specific disorder types
• Speech Disorder Symptom Analysis: Detailed symptom classification and analysis

Languages

The dataset contains English speech samples.

Dataset Structure

Data Instances

Each instance contains:

• audio: Audio file (MP3 format, 16kHz sampling rate)
• disorder_class: Binary classification (0: healthy, 1: speech_disorder)
• disorder_type: Specific type of speech disorder
• disorder_symptom: Detailed symptoms description
• transcription: Manual transcription of the speech
• age: Participant age
• gender: Participant gender
• file_name: Original filename
• split: Dataset split (test)

Data Fields

• audio: Audio feature with 16kHz sampling rate
• disorder_class: ClassLabel with names ["healthy", "speech_disorder"]
• disorder_type: String describing the type of disorder
• disorder_symptom: String describing specific symptoms
• transcription: String containing manual transcription
• age: String containing age information
• gender: String containing gender information
• file_name: String containing the original filename
• split: String indicating the dataset split

Data Splits

• test: Contains all available samples for testing/evaluation

Dataset Creation

Source Data

The dataset is created from manually labeled speech samples collected during speech-language pathology sessions.

Annotations

All annotations are manually created by trained speech-language pathologists.

Personal and Sensitive Information

This dataset contains audio recordings of human speech. All data has been collected with appropriate consent and anonymized to protect participant privacy.

Additional Information

Dataset Curators

The dataset was curated by the SAA-Lab research team.

Licensing Information

This dataset is licensed under the MIT License.

Citation Information

@misc{slp_helm_manual_labels,
  title={SLP Helm Manual Labels},
  author={SAA-Lab},
  year={2024},
  url={https://huggingface.co/datasets/SAA-Lab/SLPHelmManualLabels}
}

Contributions

Contributions to improve the dataset are welcome. Please contact the dataset curators for more information.

Usage

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("SAA-Lab/SLPHelmManualLabels", trust_remote_code=True)

# Access the test split
test_data = dataset["test"]

# Get dataset info
print(f"Test split contains {len(test_data)} samples")
print(f"Features: {test_data.features}")

Example Usage

# Access individual examples
for example in test_data:
    print(f"Disorder class: {example['disorder_class']}")
    print(f"Transcription: {example['transcription']}")
    print(f"Audio file: {example['audio']}")
    break