Théo Z. V. Champion's Whispering Wires Turn an Old Phone Into a Raspberry Pi-Powered Poet

"What if these silent wires could whisper again," the artist mused — before packing 3,000 spoken-word poems into a vintage telephone.

Gareth Halfacree
2 months agoRetro Tech / Art / HW101

Maker Théo Z. V. Champion, also known as "Rootkid," has turned a classic push-button telephone into a work of art — by replacing its innards with a Raspberry Pi Zero 2 W single-board computer and programming it to read poetry.

"A few weeks ago I visited a vintage market in London, and there was this phone," Champion explains. "Like, [a] vintage, rotary-type phone. Even though it was clearly disconnected I noticed people kept lifting its handset to their ears, on instinct. Of course [they were] met with silence, but what if these silent wires could whisper again? What if each lift of the handset delivered a voice, a fragment of life, an advice, a hardship… words from a poet."

What if an old, disconnected telephone could speak? Whispering Wires investigates just that. (📹: Rootkid)

Dubbed Whispering Wires, Champion's art piece is based on a Socotel (Société des Constructeurs de Téléphone) 63 in push-button, rather than rotary pulse-dialing, form. Rather than connecting to a plain old telephone line, as originally intended, the device is treated to an upgrade: a Raspberry Pi Zero 2 W single-board computer lives inside the housing, connecting to a Wi-Fi network to stream audio over the internet.

"I needed a huge source of narrated poems in audio format," Champion notes of said audio. "I looked around and stumbled upon the Poetry Foundation website. It has a huge library of poems, and some of them include audio narrated by the author - so I poked around the website code and was quickly able to find an open access to their database, from which I could curry poems directly."

To make the experience more seamless, Champion turned to a couple of large language models (LLMs): Whisper, which transcribes the audio with word-level timing, and OpenAI's GPT-4o-mini, which takes that transcription and attempts to identify where the actual poem — rather than introductory preamble — begins and ends. The audio is then cut to only the poem section, delivering over 3,000 spoken-word pieces with no preamble.

"I wanted the audio to sound like old phones," Champion adds, "so I did some research and found that phones back then typically used a frequency band range of 300 to 3,400Hz and a sampling frequency of 8kHz, so I wrote some code to apply this filter to the audio before playing."

The project is documented in full in the video embedded above and on Champion's YouTube channel.

Gareth Halfacree
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire: freelance@halfacree.co.uk.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles