Show HN: Inflect TTS v2+ONNX, 9M/4M text-to-speech models running in the browser

afaik controls random noise injected into the model's latent vector before decoding

same seed + same variation + same text -> bit-identical audio