GreggSpeak

Overview

GreggSpeak is a team thesis prototype for post-hearing transcription in municipal trial courts. A Raspberry Pi camera and document feeder capture handwritten Gregg shorthand notes; image processing separates lines and words, then a CNN trained on 1,480 verified word classes recognizes the shorthand. The system refines the resulting text and prepares a Transcript of Stenographic Notes (TSN)-style draft. Stenographers remain responsible for checking and correcting the output in a locally hosted web workspace before export.

Features

• Scan batches of handwritten notes using a document feeder and fixed Raspberry Pi camera.
• Preprocess page images, segment lines and words, and recognize shorthand with a MobileNetV2-based CNN deployed through TensorFlow Lite.
• Refine recognized text and prepare TSN-style drafts with editable case and transcript fields.
• Review scanned pages and recognition output, edit transcripts, and manage saved batches in a local web workspace.
• Export reviewed transcripts as PDF or editable Word documents, with text-to-speech playback.