SignTalk - A Deep Learning-Based System for Real-Time Sign Language Recognition & Voice Generation

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Gandu Lavanya, Jaligam Srilekha, Kare Ranjith Kumar, Kelim Praveen, Burra Manoj Kumar
Format: Recurso digital
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901647193538560
author Gandu Lavanya
Jaligam Srilekha
Kare Ranjith Kumar
Kelim Praveen
Burra Manoj Kumar
author_facet Gandu Lavanya
Jaligam Srilekha
Kare Ranjith Kumar
Kelim Praveen
Burra Manoj Kumar
contents Sign language is an important and expressive method of communication used by individuals who are deaf or hard of hearing. However, communication difficulties often arise when interacting with people who do not understand sign language. To overcome this issue, this research presents SignTalk AI, a real-time sign language recognition system that converts hand gestures into readable text and spoken output using artificial intelligence and computer vision techniques. The proposed system utilizes a Convolutional Neural Network (CNN) trained on the American Sign Language (ASL) Alphabet Dataset to recognize hand gestures captured through a webcam. Video frames are processed continuously, and the region containing the hand gesture is extracted and prepared before being passed to the trained model for classification. To improve prediction stability, the system uses a frame-based prediction smoothing method along with confidence threshold filtering. In addition, MediaPipe hand detection is integrated so that predictions are generated only when a hand is detected in the frame, which helps reduce noise and improves system reliability. The recognized gestures are displayed as text and can also be converted into speech using text-to-speech technology. A web application built with Flask and OpenCV provides real-time visualization of gestures, prediction tracking, sentence generation, and voice output. Experimental results show that the system performs with high recognition accuracy and stable real-time performance under normal conditions. This solution can assist speech- and hearing-impaired individuals in communication and can be extended in the future to support continuous sign interpretation and mobile-based applications.
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19594547
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle SignTalk - A Deep Learning-Based System for Real-Time Sign Language Recognition & Voice Generation
Gandu Lavanya
Jaligam Srilekha
Kare Ranjith Kumar
Kelim Praveen
Burra Manoj Kumar
Sign Language Recognition
Deep Learning
CNN
Computer Vision
ASL
Sign language is an important and expressive method of communication used by individuals who are deaf or hard of hearing. However, communication difficulties often arise when interacting with people who do not understand sign language. To overcome this issue, this research presents SignTalk AI, a real-time sign language recognition system that converts hand gestures into readable text and spoken output using artificial intelligence and computer vision techniques. The proposed system utilizes a Convolutional Neural Network (CNN) trained on the American Sign Language (ASL) Alphabet Dataset to recognize hand gestures captured through a webcam. Video frames are processed continuously, and the region containing the hand gesture is extracted and prepared before being passed to the trained model for classification. To improve prediction stability, the system uses a frame-based prediction smoothing method along with confidence threshold filtering. In addition, MediaPipe hand detection is integrated so that predictions are generated only when a hand is detected in the frame, which helps reduce noise and improves system reliability. The recognized gestures are displayed as text and can also be converted into speech using text-to-speech technology. A web application built with Flask and OpenCV provides real-time visualization of gestures, prediction tracking, sentence generation, and voice output. Experimental results show that the system performs with high recognition accuracy and stable real-time performance under normal conditions. This solution can assist speech- and hearing-impaired individuals in communication and can be extended in the future to support continuous sign interpretation and mobile-based applications.
title SignTalk - A Deep Learning-Based System for Real-Time Sign Language Recognition & Voice Generation
topic Sign Language Recognition
Deep Learning
CNN
Computer Vision
ASL
url https://doi.org/10.5281/zenodo.19594547