Published Paper Details:

Communication barriers between sign language users and non-signers continue to affect millions of individuals with hearing and speech impairments worldwide. Existing sign language translation systems frequently rely on sensor-based gloves, wearable devices, or specialized hardware, resulting in increased implementation cost, reduced portability, and complex maintenance requirements. To address these limitations, this research proposes VisionSpeak, a real-time camera-based sign language to speech translation framework integrating computer vision, deep learning, and intelligent speech synthesis technologies for affordable and accessible communication assistance. The proposed system develops an AI-driven gesture recognition framework capable of capturing, interpreting, and translating sign language gestures into audible speech using a standard webcam. The architecture consists of image acquisition, hand landmark detection, preprocessing and normalization, deep learning-based gesture classification, and real-time speech synthesis modules. OpenCV is utilized for video frame acquisition and processing, while MediaPipe performs real-time detection of 21 hand landmarks including fingertips, finger joints, and wrist positions. The extracted landmark coordinates are normalized and converted into feature vectors for intelligent gesture analysis. A TensorFlow and Keras-based neural network model is implemented for gesture classification using a custom dataset collected under varying environmental conditions and lighting scenarios. The model is trained using dense neural network layers, dropout regularization, categorical cross-entropy optimization, and cross-validation techniques to improve classification accuracy and reduce overfitting. Furthermore, the recognized gesture labels are converted into natural speech output using gTTS and pyttsx3 text-to-speech libraries, enabling seamless communication between sign language users and the general public. The proposed VisionSpeak framework is implemented using Python, TensorFlow, OpenCV, MediaPipe, and speech synthesis libraries, and evaluated using real-time gesture datasets consisting of alphabet gestures, common words, and simple conversational expressions. Experimental analysis demonstrates that the system achieves an average gesture recognition accuracy exceeding 94%, with a real-time response latency below 200 milliseconds. Performance evaluation under multiple lighting conditions further confirms the robustness, scalability, and real-time usability of the proposed system. The major contribution of this work lies in developing a low-cost, hardware-free, and AI-powered assistive communication framework capable of delivering accurate real-time sign language translation using only a webcam-based interface. The proposed solution promotes digital accessibility, inclusive communication, and intelligent human-computer interaction, while establishing a scalable foundation for future research in multimodal gesture recognition, continuous sign language translation, and AI-driven assistive technologies.