Credits

A nod to the teams whose work this computer runs.

The versions were read on the live machine on 2026-10-07.

Resemble AI

Chatterbox Turbo is the conversation voice. chatterbox-tts 0.1.7, MIT. The Perth watermark stays on: resemble-perth 1.0.1.

Audio8

The ear is their ark-asr model, 0.6B int8 ONNX, on CPU. The announcer is their TTS runtime. Thank you for both.

TEN Framework

TEN VAD 1.0.6.8 decides when someone is speaking. Apache 2.0 with additional conditions. One pitch file carries LPCNet’s BSD code, noted in their NOTICES.

WeSpeaker

WeSpeaker 0.0.1, Apache-2.0, is the voice encoder on the ear.

Silero

Silero VAD 6.2.1 is installed beside TEN, ready if the switch goes back.

PipeWire

PipeWire 1.6.2 is the room’s audio server. The microphone and the speakers meet there.

vLLM

vLLM runs both language lanes: 0.30.0 for replies, 0.27.1 for the other lane. The weight they serve is Qwen3.8-27B-NVFP4. Thank you to the vLLM team and the Qwen team.