Author uses Chatterbox TTS' zero-shot voice cloning to generate synthetic training data from a single phrase, Whisper STT to verify the generated voice sample to catch generation errors, and then uses the synthetic data set to fine-tune Piper TTS the standard way.