riker@muttnet ~ $

AI Voice Generation

published

As part of my project at https://fcc.cc/news/ I’ve been using Piper TTS to generate audio from text. Both for headlines and full article audio, in the case that the news source allows full republication. Piper works fine for a news reader but it really has a robotic feel. I just switched to Kokoro and it’s a mind-blowingly dramatic difference. (Both Piper TTS and Kokoro run locally for free.)

Same script, same encoder settings. Old TTS on top; the two new Kokoro voices below.
  • Before Piper · en_US-lessac-high 22 kHz — the engine fcc.cc/news rode for ~6 weeks
  • After Kokoro · af_heart 24 kHz — domestic & features desks
  • After Kokoro · am_michael 24 kHz — world & science desks