Household karaoke
PiKaraoke in the homelab: guests queue K-pop, Mandopop and Cantopop from their phones, and local AI models keep the library tidy.
- Status
- Running
- Timeline
- Sept 2026 to 29 Sept 2026
- Built with
- PiKaraoke, Proxmox VE, n8n, Local LLMs, OpenCC
The karaoke machine at home has good speakers, mics and effects, and a slow, locked-down Android tablet running it. The tablet has been demoted to speaker duty.
What runs now
PiKaraoke runs headless in a container on the Proxmox cluster, with the song library on the NAS. The PC by the TV shows the full-screen player and sends audio to the karaoke machine. Guests scan a QR code to search and queue songs from their phones.
The part I built
- A library worth singing from. The content is K-pop, Mandopop, Cantopop and English. There is no official Chinese KTV channel on YouTube to lean on, so candidate sources were picked by sampling what they actually publish, not by their titles.
- Songs added from phones get filed properly. A nightly n8n workflow asks two local models where each new song belongs, and only moves it when they agree. Chinese titles and artist names are built on the server with OpenCC and pypinyin, and artists keep their official spellings.