<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Lora on tech.mytrapp.com</title><link>https://tech.mytrapp.com/tags/lora/</link><description>Recent content in Lora on tech.mytrapp.com</description><generator>Hugo</generator><language>en-us</language><copyright>David Trapp</copyright><lastBuildDate>Mon, 07 Sep 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://tech.mytrapp.com/tags/lora/index.xml" rel="self" type="application/rss+xml"/><item><title>How I Gave My Kiosk a Voice Clone Trained on 213 Lines of Reach Cortana</title><link>https://tech.mytrapp.com/posts/cortana-voice-clone/</link><pubDate>Mon, 07 Sep 2026 00:00:00 +0000</pubDate><guid>https://tech.mytrapp.com/posts/cortana-voice-clone/</guid><description>The implementation guide for the voice arc of my family kiosk: finding and decoding Reach&amp;#39;s FMOD banks in my own MCC install, mapping 31,433 named streams to one speaker, gating a 213-line corpus on pitch and provenance after a male-voice contamination catch, an F5-TTS false start, the OmniVoice zero-shot spike, and a 213-line LoRA finetune merged to a standalone model.</description></item></channel></rss>