models/model_meta.json

{
  "model_name": "gcrn_fixed_4sec",
  "framework": "onnx",
  "source": "gcrn_fixed_4sec.onnx",
  "source_sha256": "9e8e0acd13126c9d5d19aae6f83f230a16c8914f100744153c78c58920d1ff4a",
  "inputs": [
    {
      "name": "stft_input",
      "shape": [
        1,
        2,
        401,
        161
      ],
      "dtype": "float32",
      "layout": "NCHW"
    }
  ],
  "outputs": [
    {
      "name": "stft_output",
      "shape": [
        1,
        2,
        401,
        161
      ],
      "dtype": "float32"
    }
  ],
  "opset": 12,
  "onnxsim_cosine_min": 1.0,
  "preprocess": {
    "description": "STFT of 16 kHz mono waveform -> [1,2,401,161] (batch, real/imag, time, freq)",
    "sample_rate": 16000,
    "duration_sec": 4.0,
    "n_fft": 320,
    "hop_length": 160,
    "win_length": 320,
    "center": true,
    "window": "hamming(sym=True)"
  },
  "notes": [
    "\u8f93\u5165\u4e3a STFT \u590d\u6570\u8c31\uff08\u5b9e\u90e8/\u865a\u90e8\uff09\uff0c\u65e0 mean/std \u5f52\u4e00\u5316"
  ]
}