{
  "context_recp": 7,
  "conv_delay": 9,
  "decoder_layers": 2,
  "encoder_conv_cache_len": 15,
  "encoder_dim": 256,
  "encoder_layers": 4,
  "feat_type": "logmel23_cummn",
  "frame_hz": 10.0,
  "full_output_dim": 12,
  "head_dim": 64,
  "hop_length": 80,
  "input_dim": 345,
  "input_names": [
    "frame",
    "enc_ret_kv",
    "enc_ret_scale",
    "enc_conv_cache",
    "dec_ret_kv",
    "dec_ret_scale",
    "top_buffer",
    "ingest",
    "decode"
  ],
  "key_dim": 64,
  "max_nspks": 12,
  "max_speakers": 10,
  "n_fft": 1024,
  "n_mels": 23,
  "num_heads": 4,
  "onnx_opset_version": 17,
  "output_names": [
    "full_logits",
    "enc_ret_kv_out",
    "enc_ret_scale_out",
    "enc_conv_cache_out",
    "dec_ret_kv_out",
    "dec_ret_scale_out",
    "top_buffer_out"
  ],
  "real_output_dim": 10,
  "sample_rate": 8000,
  "state_shapes": {
    "dec_ret_kv": [
      2,
      12,
      4,
      64,
      64
    ],
    "dec_ret_scale": [
      2,
      12,
      4
    ],
    "enc_conv_cache": [
      4,
      1,
      15,
      256
    ],
    "enc_ret_kv": [
      4,
      1,
      4,
      64,
      64
    ],
    "enc_ret_scale": [
      4,
      1,
      4
    ],
    "top_buffer": [
      1,
      19,
      256
    ]
  },
  "subsampling": 10,
  "target_sample_rate": 8000,
  "top_buffer_len": 19,
  "win_length": 200
}