model.json

{
  "model" : {
    "fm-decoder-downsampling-factor" : "1,2,4,2,1",
    "fm-decoder-num-layers" : "2,2,4,4,4",
    "fm-decoder-cnn-module-kernel" : "31,15,7,15,31",
    "fm-decoder-feedforward-dim" : 1536,
    "fm-decoder-num-heads" : 4,
    "fm-decoder-dim" : 512,
    "text-encoder-downsampling-factor" : "1",
    "text-encoder-num-layers" : "4",
    "text-encoder-feedforward-dim" : 512,
    "text-encoder-cnn-module-kernel" : 9,
    "text-encoder-num-heads" : 4,
    "text-encoder-dim" : 192,
    "query-head-dim" : 32,
    "value-head-dim" : 32,
    "pos-head-dim" : 4,
    "pos-dim" : 48,
    "time-embed-dim" : 192,
    "text-embed-dim" : 192
  },
  "feature" : {
    "sampling_rate": 24000,
    "frame_shift_ms": 256 / 24000 * 1000,
    "feat_dim": 100,
    "n_fft" : 1024,
    "hop_length" : 256
  }
}