§ 2024Computer Vision · NLPML Engineer

LipReader— 2024

Visual speech recognition is hard — most demos rely on perfectly framed faces and pre-cropped video.

lipreader

No live preview · view source on GitHub

01 / Problem

Visual speech recognition is hard — most demos rely on perfectly framed faces and pre-cropped video.

02 / Approach

Trained a LipNet model on the GRID Corpus and shipped a Tkinter desktop interface that streams real-time predictions over uploaded video.

03 / Impact

Real-time lip-reading prototype with end-to-end UI — coursework for CSE439 Machine Vision and CSE442 NLP-UI/UX.

04 / Stack

PythonTensorFlowLipNetOpenCVTkinter

05 / Links