Data from: An AI-driven, wearable, conformal ring system for real-time and user-independent sign language interpretation

Park, Jaejin 1 ; Kang, Kyowon 2 ; Yu, Ki Jun 1 ; Hwang, Dosik1; Shin, Yejee1

Published Apr 03, 2026 on Dryad. https://doi.org/10.5061/dryad.nvx0k6f5d

Data files

Apr 03, 2026 version files 313.32 MB

Added_signers.zip

20.63 MB
Raw_data.zip

292.67 MB
README.md

4.81 KB
WRSLT_code.zip

10.07 KB

Abstract

Sign language translation systems have long aimed to bridge the communication between signers and non-signers. However, preliminary systems rely on glove-type wearables or wired sensor arrays, which constrain hand movement, reduce comfort, and require non-personalized sensor positions that limit adaptability across users. Here, we introduce a wirelessly-connected, ring-type sign language translator (WRSLT) designed to overcome these limitations by enabling full finger mobility through independent sensor rings and multi-link communication. The system supports static and dynamic gesture detection using selected fingers via quantitative relevance analysis, and achieves robust user-independent performance without per-user calibration. WRSLT demonstrated high recognition accuracy on large-scale datasets comprising 100 American Sign Language and 100 International Sign Language words, achieving 88.4% and 88.9% accuracy, respectively, under unseen-user conditions (i.e., test users not included in model training). Furthermore, a custom sequential word detection framework enables sentence-level translation from continuous signing input without requiring separate training on entire sentence structures.

Dataset DOI: 10.5061/dryad.nvx0k6f5d

Description of the data and file structure

This dataset contains training and test data for a 200-word sign language recognition system based on a wearable ring-type sensor platform (WRSLT: Wearable Ring-based Sign Language Translation system). The dataset includes both American Sign Language (ASL) and International Sign Language (ISL), enabling cross-language and user-independent recognition.

The dataset is organized as follows:

Raw_data.zip
└── Raw_data/
├── Training Data/
│ ├── Only ASL/
│ ├── Only ISL/
│ └── Same ASL ISL/
├── Test Data/
│ ├── HJH_TEST_Final/
│ ├── PJH_TEST_Final/
│ └── SYJ_TEST_Final/
├── Sentence Data/
└── Sequence Changed Sentence Data/
Added_signers.zip
WRSLT_code.zip\

Training Data
The training dataset consists of 200 words (100 ASL and 100 ISL).
- Only ASL: ASL-only word data
- Only ISL: ISL-only word data
- Same ASL ISL: Semantically identical words represented in both ASL and ISL
Test Data
The test dataset contains recordings from three independent participants (HJH, PJH, SYJ), used for user-independent evaluation.
Sentence Data
Contains sequential gesture recordings representing natural phrases (e.g., “boy forget friend name”, “girl like good game”), with multiple repetitions for each sentence.
Sequence Changed Sentence Data
Contains variations of sentence sequences with altered word order, used to evaluate robustness of sequence recognition.

Files and variables

File: Raw_data.zip

Description: Test and training dataset for wearable ring-based sign language recognition.

Data format: Multichannel time-series sensor data

Each folder corresponds to a word or sentence instance

Multiple repetitions are provided for each word/sentence (e.g., name1–name5)

Variables (general description)

The dataset contains time-series signals representing finger motion and gesture dynamics captured by a wearable ring system.

Each file corresponds to one gesture instance

Signals represent temporal changes during sign execution

Data are suitable for machine learning and deep learning-based classification and sequence modeling

Additional details on data organization and naming conventions

For the Training Data and Test Data, the folder names (JJ, KW, HJH, PJH, SYJ) correspond to the initials of the participants involved in the data collection.

Within each data file, the channel naming convention is defined as follows:

L1–L5: Left hand fingers

L1: Left thumb

L2: Left index finger

L3: Left middle finger

L4: Left ring finger

L5: Left little finger

R1–R5: Right hand fingers

R1: Right thumb

R2: Right index finger

R3: Right middle finger

R4: Right ring finger

R5: Right little finger

At the lowest folder level, each .csv file (and corresponding .txt file) contains three columns representing spatial sensor measurements:

First column: x-axis values

Second column: y-axis values

Third column: z-axis values

These values correspond to the three-dimensional motion signals captured from each finger during gesture execution.

Added_signers.zip

The “Added_signers” folder contains additional data collected from unseen participants.
This dataset is used as a test set for evaluating user-independent performance, allowing assessment of the model’s generalization capability to new users who were not included in the training process.

Code/software

The file WRSLT_code.zip contains the same source code as the GitHub repository referenced in this study.

This code includes data preprocessing, model implementation, and evaluation scripts used for sign language recognition.

The GitHub repository is available at: https://github.com/yejees/WRSLT

The code is provided in both Dryad (as WRSLT_code.zip) and GitHub to ensure long-term accessibility and reproducibility.

To run the code in the WRSLT-main folder, please install the required dependencies listed below:

numpy>=1.21.0
torch>=1.10.0
scipy>=1.7.0
tqdm>=4.60.0
argparse
matplotlib>=3.4.0
seaborn>=0.11.0
pandas>=1.3.0
scikit-learn>=1.0.0
natsort>=8.0.0
git+https://github.com/openai/CLIP.git

You can install all dependencies using:

pip install -r requirements.txt

After installing the dependencies, the training script can be executed with the following command:

python train.py