A DIY Transcription Appliance for the Hard of Hearing

Hacking hardware to make lights blink or repair something that has broken over the course of time is all great stuff to do. But when we can use our tech skills to genuinely improve the lives of those around us, that’s when engineering becomes more important than just a hobby. Andy Massey just completed a project in this latter category that makes life much easier for his elderly father. His father is deaf and struggles to understand what others are saying. But Massey’s live transcription device lets him read everything as it happens.

The build looks like nothing more than a tablet sitting beside a landline phone. But it’s actually a purpose-built assistive appliance centered on a Raspberry Pi 5 with 8 GB of RAM and a 10-inch 1280×800 touchscreen display. The Pi runs a custom Python application that listens to conversations and converts speech into large, readable captions in near-real-time.

For phone calls, Massey tapped into the landline using a small USB telephone recorder connected through an RJ-11 splitter. This captures audio directly from the call without needing speakerphone volume cranked to uncomfortable levels. For in-room speech — from visitors, caregivers, or family members — a USB conference microphone continuously listens.

Software decides which audio source matters at any given time. When a call begins, the system automatically switches to the phone input and displays a small phone icon on the screen. When the call ends and silence returns, it reverts to the room microphone after about ten seconds.

Speech recognition is handled by multiple engines for reliability. Under normal conditions, the device sends audio to a cloud transcription service for high-accuracy captions that appear in a fraction of a second. If the internet connection drops, the program automatically falls back to offline models running locally — first a Whisper-based recognizer and then Vosk — ensuring captions still appear even during outages.

As far as the interface is concerned, captions are shown in very large text with selectable color schemes for contrast, and the touchscreen allows scrolling through recent dialogue. When nobody is talking, the display transforms into a split-flap style flip clock that dims automatically overnight.

The Pi boots directly into the captioning software as a background service, and watchdog timers restart the application, the display manager, or even the network if something locks up. This makes the unit behave less like a computer and more like a household appliance, which is perfect for a user who is not equipped to deal with tech issues that may arise.

For more details on the build, take a look at the GitHub repository. Everything has been released under an MIT license.