I Gave a Pumpkin Super Intelligence

Highlights

• A local voice model runs on a Raspberry Pi using a speech-to-text model called parakeet.

• A remote LLM named deep seek flash processes text inputs because it provides the fastest response time under one second.

• Eleven labs flash converts text responses back into speech for the final audio playback.

• Poor soldering caused the perma proto board pins to touch the cooler and short-circuit the speaker until a piece of card wedged between them fixed the issue.

Timeline

1. 00:00 - 01:40

   Hardware Planning and Design

   The project starts with the goal of building a functional smart speaker inside a pumpkin. The design uses a Raspberry Pi connected to a hat with an amplifier, speaker, two LEDs, and a microphone. Software components include a trigger word listener, a local speech-to-text model, and a remote language model for processing.

   • Standard smart speakers often fail to answer basic custom questions.
   • A local voice model on a Raspberry Pi ensures fast replies.
   • External tools connect through a remote LLM using location and web access.
2. 01:40 - 05:24

   Soldering Components and Building the Circuit

   Building the hardware involves learning how to solder from scratch using online videos and a practice kit. Soldering the LED light rings takes hours because fragile wires keep snapping. Once the perma proto hat has all components attached, it mounts directly onto the raspberry pi along with an active cooler, microphone, and speaker.

   • A practice soldering kit helps build beginner skills before handling expensive parts.
   • Connecting two led light rings requires splicing and soldering multiple cables into small holes.
   • A perma proto board holds forty pins, an amplifier strip, and all wiring before connecting to the raspberry pi.
3. 05:24 - 06:54

   Software Setup and Debugging

   Software configuration relies on Python to handle voice inputs and route them to deep seek flash for rapid text generation. Eleven labs flash turns the output text back into speech for the speaker. Debugging is required when the speaker stops working due to the proto board pins making contact with the cooling unit, which gets resolved by inserting a piece of card as insulation.

   • Python runs text-to-speech processing via sherpa onnx and a parakeet model.
   • Deep seek flash delivers the fastest response times under one second compared to gemini flash and haiku.
   • A short circuit occurs when poorly soldered pins touch the active cooler.
4. 06:54 - 10:20

   Carving and Final Testing

   The final assembly step involves carving a face into the pumpkin and lining the interior with cling film to protect the raspberry pi from moisture and conductivity issues. After placing the hardware inside and connecting power, the smart speaker successfully engages in dialogue, tells jokes, and creates weekend plans.

   • Cling film lines the inside of the pumpkin to prevent short circuits against the electronics.
   • The assembled smart speaker successfully answers conversational prompts and suggests local event itineraries.
   • Total parts cost more than a retail smart speaker.