{
  "model": "ToneColorConverter_E2E.mlpackage",
  "inputs": {
    "y": "[1, 513, T] float32 — linear STFT magnitude (same layout as PyTorch OpenVoice)",
    "y_lengths": "[1] int32",
    "sid_src": "[1, 256, 1] float32 — source tone embedding",
    "sid_tgt": "[1, 256, 1] float32 — target tone embedding"
  },
  "outputs": {
    "audio_out": "[1, 1, T_audio] float32"
  },
  "notes": [
    "tau=0 固定（後験からのサンプリングなし）。",
    "5 分割版と異なり中間 z を取り出せないが呼び出しは1回で済む。"
  ]
}