Maker · Recent Project

AIY Voice Kit (V1) — Final Build Report

Bought in 2017, assembled in 2026 — nine years on the shelf, working end to end

Status: Complete. Google's AIY Voice Kit converts speech to text via Google Cloud, now fully working end to end — mic → cloud transcription → recognized command. The official repo has been archived since Feb 2023, so getting there meant a handful of real fixes rather than following the original instructions as written.

The short version

Repo archived by Google since Feb 2023 — no more official fixes, but the last release (2021-04-02) still works with a handful of patches.
First hardware mixup: assumed the spare board was a Raspberry Pi 3. It was actually an original 2014 Pi 1 Model B+ — no WiFi hardware at all. Solved by switching to a Pi 4.
Pi 4 needs its own power supply (USB-C, 5V/3A) — the Pi 3's supply is physically incompatible and won't work.
WiFi config file had a typo from manual entry — fixed with wpa_passphrase instead of hand-editing.
The Pi's bundled browser is too old for Google's modern sign-in — worked around by authenticating via SSH/terminal instead of the desktop GUI.
Hit a real code bug: AttributeError: module 'google.cloud.speech' has no attribute 'types' — caused by the google-cloud-speech Python library moving on from the old .types structure since the kit's software was last updated. Fixed with a targeted sed patch to cloudspeech.py.
Google Cloud now requires a billing account (card on file) even for free-tier usage — a change since the kit's original 2017 setup instructions were written. Free tier (60 min/month speech recognition + $300 new-account credit) comfortably covers hobby use.
Final test (cloudspeech_demo.py): live transcription worked cleanly, including LED/action mapping for phrases like "turn on the light" and "goodbye."

What this project actually demonstrated

Diagnosing hardware identity issues methodically (cat /proc/cpuinfo) rather than assuming
Recognizing when a symptom (undervoltage, WiFi failure) has a root cause elsewhere (power supply) rather than treating each surface-level failure independently
Patching a real, documented open-source library incompatibility rather than reflashing/reinstalling as a first resort
Navigating a service (Google Cloud) that has changed its account/billing requirements since the original documentation was written, without official kit support to fall back on

Full step-by-step details, including every command run and why, are in the setup guide:

→ Read the full setup guide

What's next

Next in the series: the AIY Vision Kit, followed by integration between the two, and eventually a Pi 4-based, locally hosted version — moving away from Google Cloud's speech API entirely.