voice-lab

Unified voice workbench for the human-AI voice collaboration loop. Combines TTS quality evaluation, RVC parameter sweep (formerly hum-lab), PipeWire recording, follow-along sessions, and voice lesson package creation.

What Horus Receives

A voice lesson package per session:

• SRT — text + timing (what was said, when)
• stems — original artist vocals (from create-stems)
• .pt file — current PersonaPlex voice anchor
• graham.wav — human performance aligned to the same content
• prompt.md — learning guidance (what register, what emotion, what to focus on)
• training_data.jsonl — Qwen3-TTS fine-tuning format

Quick Start